• 专利附录
  • 专利说明
  • 权利要求
  • 类似技术
  • 同族专利
  • 引用文献
  • 法律状态
  • 优先权
    • 专利摘要:
    The invention relates to a method for combining a first optical character recognition (OCR) (12) and a second OCR (100). The first OCR (12) operates first on a string image (11). Its output (13) (first identified characters, character positions and character likelihood parameters) is used to generate a first graph (16). Segmentation points related to the positions of the first identified characters (14) are used as input by the second OCR (100) by performing combined segmentation and classification on the character string image (11). The output (17) (second identified characters, character positions and character likelihood parameters) of the second OCR (100) is used to update (20) the first graph (16) to generate a second graph (21). ) which combines the output (13) of the first OCR (12) with the output (17) of the second OCR (100). Decision models are then used to modify (22) the path weights in the second graph (21) to generate a third graph (23). A better path is determined (24) on the third graph (23) to obtain identification (25) of the characters present in the character string image (11).
    公开号:BE1024194A9
    申请号:E20165773
    申请日:2016-10-14
    公开日:2017-12-19
    发明作者:Frédéric Collet；Jordi Hautot；Michel Dauw；Meulenaere Pierre De
    申请人:Iris Sa；
    IPC主号:
    专利说明:

    Method for authenticating a character in a digital image
    Technical area
    The present invention provides methods and programs for identifying characters in a digital image.
    Background technique
    Optical Character Recognition (OCR) methods are known that convert text present in an image into machine readable code.
    U.S. Patent No. 5,519,788 discloses a method of implementing a weighted voting scheme for accurately reading and recognizing characters in a scanned image. A plurality of optical character recognition processors scan the image and read the same image characters. Each OCR processor delivers a reported character corresponding to each character read. For a particular character reading, the characters reported by each OCR processor are grouped into a set of candidate characters. For each character candidate, weighting is generated according to a confusion matrix that stores the probabilities of a particular OCR to precisely identify the characters. The weights are then compared to determine which character candidate to issue.
    Such a process has several limitations. First, since the OCR processors are operated in parallel, the character that would not be recognized by any of the independently taken OCR processors can not be recognized by the process as a whole.
    Secondly, a preprocessor must quantify the strengths and weaknesses of the OCR processors to generate the confusion matrix cells that contain probabilities that the character read by an OCR processor is the character reported by the OCR processor. This step can take a long time. In addition, if this step is not well executed, by the fact that the training set used for this purpose is not suitable for a given type of character, a probability may be low for a character that is actually good. recognized and the OCR process can provide worse results than your OCR processors taken independently,
    Description of the invention
    An object of the present invention is to provide an improved method for applying multi-character identification methods to a digital image in order to obtain better and / or faster identification results. This object is achieved according to the invention by a method of selecting candidate characters in a method of identifying characters of a digital image, the method comprising the steps of: a) applying a first character identification process for determining first character candidates and a list of segmentation points of the first character candidates, b) generating a list of character widths corresponding to a segmentation point of between the list of segmentation points, c) determining a part of the digital image corresponding to the segmentation point and a character width of the character width list, d) applying a character classification method on the part of the digital image to obtain a hypothesis ID of a possibly character present in the part of the digital image and a likelihood parameter that relates to the vr the likelihood that the hypothesis ID is correct and e) selecting the hypothesis ID as the second candidate character in the digital image if the likelihood parameter fulfills a first predetermined criterion.
    In the method according to the invention, the segmentation points determined from the first character identification process (first OCR) are used as input for a combined segmentation and classification process (which comprises steps b, c, d and e), which is a second OCR. As a result, the second OCR is able to perform better segmentation than if it operated alone without the input of the first OCR. This allows the second OCR to use segmentation points that it would not have determined alone and thus to recognize certain characters, for example Asian characters, it would not have recognized alone. In addition, the second OCR works faster with this entry of the first OCR because its initial estimate of the next character's width is better than without this entry.
    In embodiments of the invention, the method further comprises a step f) updating the list of segmentation points with another segmentation point determined on the basis of the portion of the digital image corresponding to the second selected candidate character. This other segmentation point is expected to be useful for the character following the character that has just been identified.
    In one embodiment of the invention, step f) comprises: checking whether the other segmentation point is already in the list of segmentation points and the addition of the other segmentation point to the list of segmentation points if the other segmentation point is not already present in the segmentation point list.
    In one embodiment of the invention, in step b), the list of character widths is generated based on at least the list of segmentation points of the first character candidates. The list of character widths can therefore be generated quickly and reliably because its determination is based on data from the first OCR.
    Advantageously, steps b, c, d and e are performed for each segmentation point of the list of segmentation points.
    In one embodiment of the invention, steps c, d and e are repeated for another character width of the character width list if the likelihood parameter does not satisfy the first predetermined criterion. If the likelihood parameter does not satisfy the first predetermined criterion, it indicates that the hypothesis ID may not be correct. It is therefore better to consider another part of the digital image starting at the same segmentation point and consequently covering at least in part the same connected components.
    In one embodiment of the invention, steps c, d and e are repeated for another character width of the character width list if the likelihood parameter satisfies the first predetermined criterion and does not satisfy a second criterion. predetermined.
    In one embodiment of the invention, the method further comprises the steps of: generating a data structure from the first character candidates and their segmentation points and updating the data structure with the second character candidate selected and a next segmentation point calculated from the character width of the selected second character candidate and the segmentation point used to determine the portion of the digital image corresponding to the selected second character candidate.
    A first purpose of the data structure is to combine the results of the character classification method with the results of the first character identification process.
    In one embodiment of the invention, step a) comprises determining first likelihood parameters of the first candidate characters providing an indication of the chance that the candidate character is correct, the data structure comprises the first likelihood parameters. first candidate characters and the method further comprises the steps of: changing the scale of the first likelihood parameters of the first candidate characters and / or the likelihood parameters of the second candidate characters to make them comparable to each other and updating the data structure with the likelihood parameters of the second candidate characters.
    The likelihood parameters increase the accuracy of the process.
    Preferably, the data structure is a graph in which the segmentation points are represented by vertices and the candidate characters are represented by edges between the vertices.
    Advantageously, the method comprises the step of applying a decision modeling method on the updated data structure. Decision models improve the accuracy of the process.
    In one embodiment of the invention, the method further comprises the steps of: determining a better path in the updated data structure and selecting candidate characters corresponding to the edges of the best path as characters of the digital image ,
    In this way, the first OCR, the second OCR and the decision models are all taken into account to directly determine the identification of the characters.
    Preferably, the step of applying a decision modeling method comprises the advantage of the candidate characters that have the first likelihood parameters of the first candidate characters and the second likelihood parameters of the second candidate characters satisfying a third predetermined criterion.
    In one embodiment of the invention, the first character identification method provides first character candidates for characters of a first type, wherein the character classification method provides second character candidates for a second type, and the decision modeling method includes advancing the first type of candidate candidates if the candidate characters are provided by the first character identification process and favoring the second type character candidates if the candidate characters are provided by the classification process.
    In one embodiment of the invention, the first character identification method comprises a decision modeling method.
    In one embodiment of the invention, the first character identification process provides positions of first character candidates and further comprises a step of determining the list of segmentation points of the first character candidates from the positions of the first characters. candidate characters. If the segmentation points are not provided by the first character identification process, they can be determined from the first candidate positions by combined segmentation and classification or by an independent step performed preferably between steps a) and b) and possibly between steps b) and c).
    Another object of the present invention is to provide a method of identifying characters in a digital image.
    This object is achieved according to the invention by a method comprising: a) applying a first character identification method for determining first character candidates and first segmentation points corresponding to the first character candidates; b) generating of a first graph on the basis of the first candidate characters and the first segmentation points, in which the first segmentation points are represented by vertices and the first candidates characters by edges, c) the application of a second method character identification on at least a portion of the digital image to generate at least a second character candidate, d) updating the graph by adding the at least one second character candidate to the first graph and e) the selection in the updated graph of candidate characters as characters of the digital image.
    In this method, the graph, when generated, comprises as vertices the segmentation points coming from the first OCR and as edges the 1Ds coming from the first OCR. The graph is then updated with new vertices and new edges from the second OCR. During the selection of candidate characters (step e), all edges are considered and none of them is a priori discarded. There is no need to test the OCRs against each other to determine their weaknesses, since all IDs from all OCRs are considered even if a decision rule favoring certain edges according to some conditions could be introduced into the graph. There is no need for a structure like confusion matrix either.
    Preferably, the second character identification method generates at least one second segmentation point and the at least one second segmentation point is added to the graph in step d).
    In one embodiment of the invention, the first character identification method provides first likelihood parameters providing a likelihood that the first character candidates are correct, the second character identification method provides second likelihood parameters providing a likelihood that the at least one second character candidate is correct, and the method further comprises the steps of: changing the scale of the first likelihood parameters and / or the second likelihood parameters to make them comparable to each other; other, add the first likelihood parameters in the first graph and add the second likelihood parameters in the updated graph.
    Advantageously, the method comprises the step of applying a decision modeling method to the updated graph.
    In one embodiment of the invention, the decision modeling method includes a rule that favors candidate characters that have the first likelihood parameter and second likelihood parameters that satisfy a third predetermined criterion.
    In embodiments of the invention, some of the first and second character candidates are characters of a first or second type and the decision moderation method includes a rule which favors the candidate characters corresponding to the first type of characters if these candidate characters were determined by the first character identification method and favors candidate characters corresponding to the second type of characters if these candidate characters were determined by the second character identification method.
    Preferably, the second character identification method uses as input, in step c), the first segmentation points from the first character identification method. With this method, the second character identification method is able to perform a better segmentation than if it operated alone,
    Another object of the present invention is to provide a computer program that applies better multi-character identification processes to a digital image for better and / or faster identification results.
    This object is achieved according to the invention by a computer readable non-transitory medium storing a program causing a computer to execute a method for identifying characters in a digital image, the method comprising the steps of: a) applying a first method of identification of characters to determine first candidate characters and a list of segmentation points of the first candidate characters, b) generate a list of character widths corresponding to a segmentation point of the list of segmentation points, c) determine a part of of the digital image corresponding to the segmentation point provided by the position of a first candidate initial character and to a character width from the list of character widths; d) to apply a character classification method on the part of the character digital image to obtain a hypothesis ID of a prospective character present in the part of the Digital Image and a likelihood parameter which relates to the likelihood that the hypothesis ID is correct and e) select the hypothesis ID as a character in the digital image if the likelihood parameter responds to a first predetermined criterion.
    The method according to the invention is preferably designed to run on a computer.
    BRIEF DESCRIPTION OF THE DRAWINGS The invention will be better understood by means of the following description and the appended figures.
    Fig. 1 illustrates a block diagram of an OCR method according to the invention.
    Fig. 2 illustrates an example of an image of a character string with segmentation points.
    Fig. 3 illustrates an example of a graph corresponding to the image and the segmentation points of FIG. 2.
    Fig. 4 illustrates a combined segmentation and classification functional diagram according to an embodiment of the invention.
    Fig. 5 illustrates an update of a list of segmentation points according to one embodiment of the invention.
    Figs. 6a to 6e illustrate a very simple example of the combined segmentation and classification according to the invention.
    Fig. 7 illustrates a graph at the end of the combined segmentation and classification illustrated in FIG. 6.
    Embodiments of the invention
    The present invention will be described with reference to particular embodiments and with reference to certain drawings, but the invention is not limited thereto, but only by the claims. The drawings described are only schematic and are not limiting. In the drawings, the size of some of the elements may be exaggerated and not scaled for illustrative purposes. The dimensions and relative dimensions therefore do not necessarily correspond to actual reductions to practice the invention.
    In addition, the terms first, second, third, etc. in the description and the claims are used to distinguish between similar elements and not necessarily to describe a sequential or chronological order. The terms are interchangeable under appropriate circumstances and the embodiments of the invention may operate in other sequences than those described or illustrated herein.
    In addition, the terms above, below, on, under and the like in the description and claims are used for descriptive purposes and not necessarily for describing relative positions. The terms so used are interchangeable under appropriate circumstances and the embodiments of the invention described herein may operate in other orientations than those described or illustrated herein.
    In addition, the various embodiments, although referred to as "preferred", are intended to be examples of modes in which the invention can be implemented rather than limiting the scope of the invention.
    The term "comprising" used in the claims should not be construed as limited to the elements or steps enumerated thereafter; it does not exclude other elements or other stages. It must be interpreted as specifying the presence of the characteristics, integers, steps or components referred to, but does not preclude the presence or addition of one or more other characteristics, integers, steps or components or their groups. As a result, the scope of the phrase "a device comprising A and B" should not be limited to devices consisting solely of components A and B rather than the present invention, the only listed components of the device are A and B and, further, the claim must be interpreted as including equivalents of these components.
    The term "character," as used herein, refers to a symbol or sign used in writing such as a grapheme, a logogram, an alphabetical letter, a lettering, a numerical digit, or a punctuation mark .
    The terms "identification, identifier, and ID," as used herein, refer to one or more characters recognized in machine-readable code to obtain an explorable string of characters.
    The term "position", as used herein with reference to a particular position, refers to data that makes it possible to locate the character. For example, the position can be provided by the coordinates of the pixels of a bounding box that surrounds the character. The term "segmentation point" as used herein refers to the point where there is a boundary between the characters. For example, a segmentation point of a character can be provided by the leftmost black pixel of a character boundary box or by the leftmost black pixel of the character. The term "segmentation portion" as used herein refers to a portion of an image of a character string that undergoes processes to determine whether it represents a character, a group of characters, a pattern, etc. The leftmost point of a segmentation part is preferably a segmentation point.
    The term "classification" as used herein refers to the generation of at least one hypothesis on the identification of one or more characters. Each hypothesis of identification or hypothesis ID is associated with a parameter of likelihood. The term "optical character recognition (OCR)" as used herein refers to any conversion into text present in a computer readable code image. An OCR can be very complex and include, for example, decision models, just as it can be a relatively simple one-character classification.
    The terms "candidates, hypotheses, 1D hypotheses or identification hypotheses", as used here, refer to other possible solutions for the identification of a character or group of characters. often related to a likelihood parameter. The term "likelihood parameter," as used herein, refers to a parameter that has a value that provides an indication of likelihood that the identification hypothesis, your 1D hypotheses, the character candidate, etc. ., is or are correct, that is, the segmentation image or segment that is being classified is actually the character or group of characters in the idenfification assumption. A likelihood parameter may be, for example, a probability or a weighting. Likelihood parameters from multiple OCRs can be at different scales, in which case a conversion can be used to place them on the same scale and make them comparable. The term "data structure" as used herein refers to an entity comprising data.
    The term "graph" as used herein refers to a data structure including vertices and edges.
    As used herein, the term "connected component" of a black and white image is meant to refer to a group of black pixels that are connected to each other by black pixels.
    Fig. 1 is a block diagram of an OCR method 10 according to the invention.
    An image 11 of a character string is taken as input by a first OCR 12. The image 11 of a character string is preferably a black and white digital image. The first OCR 12 processes the information contained in the image and provides as outputs of the first character candidates 13 with segmentation points 14 of the characters of the image and preferably weightings 18 of the characters, the weightings being likelihood parameters. The first candidate characters 13 with the segmentation points 14 of the characters in the image may be a sequence of candidate characters with the segmentation points 14 provided by the positions of the characters, possibly with related weights, or it may be a plurality of possible sequences of candidate characters with the positions of the characters possibly with related weights. The segmentation points 14 of the first candidate characters 13 may be called first segmentation points 14. The plurality of segmentation points 14 may be considered as a list of segmentation points 14.
    The first OCR may or may not use decision models.
    The first character candidates 13, with their segmentation points 14 and their weightings 18, are used as input into a generation of a first graph, which delivers a first graph 16. If necessary, the generation of the first graph determines first Segmentation points from the candidate character segmentation points 14, If necessary, the generation of the first graph converts the weights 18 into first likelihood parameters to make them correspond to another scale of likelihood parameters. In the first graph 16, the first segmentation points are represented in the form of vertices, the first character candidates being represented as edges and the first likelihood parameters of the first character candidates are represented as corresponding edge values.
    A graph typically starts with a first vertex that corresponds to an initial segmentation point, which is the leftmost segmentation point of the image of character string 11. The first vertex is connected to another vertex that corresponds to at another segmentation point by an edge corresponding to the identified character between the initial point and the other segmentation points. The graph continues with a vertex for each segmentation point and an edge for each character between the segmentation points. The character likelihood parameters are represented by edge values at the corresponding edges. The graph ends with a vertex corresponding to a point on the right of the first character in the image of the character string 11. This point on the right of the last character of the image of the character string 11 is preferably not a segmentation point. Other types of data structures than a graph can be used to order OCR outputs without departing from the scope of the invention.
    Fig. 2 illustrates an image of a string of characters 11. In this example 201, the first OCR 12 has been identified by a b starting from a segmentation point 202 with a likelihood parameter of 80% and a starting t of a segmentation point 203 with a likelihood parameter of 95%,
    Fig. 3 illustrates the first graph 301 generated in step 15 from the output 13 of the first OCR 12. The vertex 302 corresponds to the segmentation point 202. The edge 304 corresponds to the letter b and the associated parameter of 80% 305 is also included in the first graph 301. The vertex 303 corresponds to the segmentation point 203. The edge 306 corresponds to the letter t and the associated parameter of 95% 307 is also included in the first graph 301.
    Referring now to FIG. 1, the first segmentation point 14 and the image of the character string 11 are used as input for the combined segmentation and classification 100, in one embodiment of the invention, wherein the positions are delivered by the first OCR 12 instead of the first segmentation points 14, the first segmentation points 14 are determined at another stage of the process 10, for example, during the combined segmentation and classification 100. For example, if the first OCR 12 delivers the limitation boxes of the first characters, the segmentation points can be determined as the points of the leftmost black pixel in each bounding box.
    The combined segmentation and classification 100 constitutes a second OCR and will be described later with reference to FIG. 4. The combined segmentation and classification 100 preferably generate at least once for each character in the image of the character string 11 at least one second character candidate with its position and preferably a probability of the second candidate character 17. The segmentation point 121 which provided the hypothesis ID 127a which was chosen as the second candidate character 17 may be used instead of or in addition to the position of the second candidate character 17. This segmentation point 121 may be referred to as the second point The probability of the second character candidate is a second parameter of likelihood.
    In one embodiment of the invention, there may be one or more second candidate characters each having its corresponding position and preferably each having its probability determined by the combined segmentation and classification 100 for each character of the image of the string of characters 11.
    The second character candidate, its position and its probability 17 are, each time they are generated, used as an input for an update of the graph which updates the first graph 16. the end of the segmentation and the classification in combination 100, the global output of the updates 20 of the graph is a second graph 21 which combines the output of the first OCR 12 and the output of the combined segmentation and classification 100. If necessary, the update of the graph 20 converts the probabilities into second likelihood parameters having the same scale as the first likelihood parameters to make them comparable. Updating the graph adds a new edge for each second ID. Updating the graph adds a new vertex at the end of each new edge that does not end on a vertex already present in the graph.
    The second graph 21 includes an edge for each of the first IDs and for each of the second IDs associated with their respective likelihood parameter, and a vertex for each of the first segmentation points and for each of the second segmentation points. An example of the second graph is illustrated in FIG. 7 and will be described later. The edges and / or vertices of the second graph 21 may have a label that indicates whether they have been generated from the first OCR or from the combined segmentation and classification 100.
    The second graph 21 is then used as input to an optional decision model addition step 22, which generates a third decision graph 23. Step 22 can add decision models to the graph by using a weighted finite state transducer (WFST) based algorithm. This step 22 is in fact an application of a decision modeling method. The addition of decision models 22 modifies the edge likelihood parameters, that is to say the characters or the group of edges, that is to say groups of characters in order to favor those which are a priori most likely, for example because of the context. Decision models modify the weightings of the trips. Decision models are preferably related to the contextual decision. Decision models greatly improve the accuracy of identification.
    Decision models can involve bigrams, typographic metrics, word lists such as dictionaries, n-grams of characters, punctuation rules, and spacing rules.
    A first type of decision model is a linguistic model. If the word "ornate" is present in the image of the string to be identified, the first or the second OCR can for example find the word "ornate" and the word "ornate" as ID with similar likelihood parameters of the that the letters rn taken together resemble the letter m. A linguistic model using a dictionary is able to detect that the word "ornate" does not exist, while the word "ornate" exists.
    In one embodiment of the present invention, the language model uses a model of n-grams. If the word "TRESMEUR" is present in the image of the character string 11, the first or the second OCR can for example find the word "TRESMEUR" and the word "TRESMEUR" as ID with similar likelihood parameters of the fact that the letters "S" may look like the letter "5" in a printed text. A linguistic model using a bi-gram model (n-gram with n = 2) would prefer "TRESMEUR" if "ES" and "SM" had better probabilities of occurrence than "E5" and "5M".
    Another type of model used in one embodiment of the present invention is a typographic model. If the word "Loguivy" is present in the image of the string of characters 11, the first or the second OCR can for example find the word "Loguivy" and the word "Loguiw" as ID with similar likelihood parameters of the fact that the letters "y" may look like the letter "v" in printed text. A typographic model using font metrics would prefer "Loguivy" because the background position of the final character is more likely to be the lower position of a "y" (in its model) than a "v",
    In one embodiment of the present invention, the typographic model considers the position of the character in the image to check whether sizes and positions are expected or not.
    In a decision model involving punctuation and spacing rules, certain combinations of punctuation and / or spaces are disadvantaged.
    In one embodiment of the invention, a decision model is added in step 22 that favors the IDs that have the first likelihood parameter and the second likelihood parameter that satisfies a predetermined criterion. In fact, it is expected that if the same ID is discovered by the first OCR 12 and the combined segmentation and classification 100, ND is more credible. Therefore, if for an ID, both the first likelihood parameter and the second likelihood parameter are above a threshold, for example, 90%, ND can be favored.
    In one embodiment of the invention, a decision model is added in step 22, which is equivalent to an OCR vote because it favors the first OCR in some cases and the second OCR in other cases. For example, if the first OCR is known to be extremely accurate for a first type of characters, such as katakana characters, while the combined segmentation and classification 100 are known to be extremely accurate for a second type of character, such as kanji , a decision model can be added which favors the IDs corresponding to the first type of characters (katakana) if these IDs were determined by the first OCR and favors the IDs corresponding to the second type of characters (kanji) if they were determined by combined segmentation and classification 100.
    A step of determining the best path 24 is then performed on the third graph 23 to determine the ID 25 characters present in the image 11 of the string. The determination of the best path 24 determines, between the possible paths defined by the edges joined by vertices in the third paragraph 23 and taking into account the likelihood parameters associated with the edges, the path, that is to say the ID sequence, with the highest probability of matching image 11 of the string. Determining the best path 24 may use a Weighted Finite State Transducer (WFST) based decision algorithm that optimizes path weighting to find the best solution.
    Fig. 4 illustrates a block diagram for combined segmentation and classification 100 according to one embodiment of the invention.
    This block diagram comprises a stack 102 which is a list of segmentation points or a data structure comprising segmentation points. The first segmentation points 14 are used as input by a step 101 of creation of the stack. During this step 101, the initial data of the stack is determined. The first candidate characters 13 could also be used by step 101. Preferably, before this step 101, the stack 102 does not exist or is empty. In one embodiment of the invention, during this step 101, the first segmentation points 14 are placed in the stack 102.
    The combined segmentation and classification 100 begin to iterate on the stacking segmentation points with a segmentation point index i which is set equal to a first value, for example 1. A corresponding segmentation point 121 from the stack 102 is considered. The first time a segmentation point 121 is considered in the combined segmentation and classification 100, the segmentation point 121 is preferably one of the first segmentation points 14 that has been determined by the first OCR. The segmentation point 121 is preferably removed from the stack 102 so that the stack 102 contains only the segmentation points that still need to be considered. The segmentation point 121 is preferably placed in a list of segmentation points. already treated.
    Then, the combined segmentation and classification 100 determines in step 130 a list of widths 131. Preferably, this determination 130 does not use as input a list of widths determined for another segmentation point. In one embodiment of the invention, the list 131 of widths is generated in step 130 as described in US Patent Application No. 14/254 096 which is incorporated herein by reference, wherein the first Segmentation points 14 are used as input. The image of the character string 11 and / or the first character candidates can also be used as inputs. In one embodiment of the invention, the determination 130 uses a database 132 storing reference character widths, storing an average width and possibly storing other character statistics. The determination 130 is preferably based on this average width provided by the database 132 and the size or sizes of the component or components connected around the segmentation point i, where the size or sizes are provided by the image 11 of the chain of data. The determination 130 is furthermore based on this average width provided by the database 132 and the size or sizes of the connected component (s) according to the segmentation point i, the size or sizes being provided by the image 11 of the string of characters.
    In one embodiment of the invention, the determination 130 estimates, starting from the first segmentation points 14, from the cover (s) of the component (s) connected around the segmentation point 121 and from the database. 132, the width for the character following the segmentation point 121.
    The 131 list of widths is preferably ordered from the most likely character width to the least likely character width, as described in US Patent Application No. 14/254 096.
    Then, the combined segmentation and classification 100 begin to iterate over the widths of the width list 131, a width index J being set equal to a first value, for example 1. A width 124 from the width list 131 is considered. The combination of the segmentation point i and the width j provides a segmentation portion 125 which is a part of the image of the character string 11. A character classification 126 is performed on the segmentation portion 125. The classification of characters 126 is preferably a classification of unique characters. The character classification 126 may comprise a character classification designed for a first type of character, for example Asian characters, and a character classification designed for a second type of character, for example Latin characters. Such a character classification is described in US Patent Application No. 14/299 205 which is incorporated herein by reference.
    The character classification 126 provides a hypothesis ID 127a with a likelihood of error Perr 127b. The hypothesis ID 127a is a hypothesis on the character possibly present on the segmentation part 125. The likelihood of the error Perr 127b is a likelihood parameter. The likelihood of the error Ρ @ π 127b is a percentage that decreases with the probability that the classification 126 correctly identified the content of the segmentation portion 125. The probability that the classification 126 correctly identified the content of the segmentation portion 125 is equal to 1 ~ Perr. Both Perr and the probability 1-Perr can be referred to as the "second likelihood parameter" because they both provide an indication of the likelihood that the identification hypothesis 127a is correct for the character classification 126.
    The combined segmentation-classification 100 then verifies, at step 128, whether the likelihood of the Perr error responds to a first predetermined criterion which is that Perr is smaller than a threshold to have a high likelihood of error Therr · Therr is preferably chosen close to 100%, for example equal to 99.9%. Thus, the first predetermined criterion rules out assumptions that are really implausible. If the first predetermined criterion is not satisfied, the index J is incremented and a next width from the list of widths 131 is considered.
    If the first predetermined criterion is satisfied, it indicates that the hypothesis ID could have a chance of being correct. If the first predetermined criterion is satisfied, an update 110 of the stack is performed. This update 110 of the stack is described below with reference to FIG.
    If the first predetermined criterion is satisfied, an update 150 of the average list is performed and the updated average width replaces the average width in the database 132. In one embodiment of the invention, the update Day 150 of the average width is performed as described in U.S. Patent Application No. 14 / 254,096 which is incorporated herein by reference.
    In one embodiment of the invention, the update 150 of the average width is followed by the determination 130 of the width list which generates the list of widths 131 to be used for the same segmentation point i in the next iteration. widths by combined segmentation and classification 100.
    If the first predetermined criterion is satisfied, the second character candidate, its position and its probability 17 are added to the graph during the corresponding update of the graph (Fig. 1). The second character candidate, its position and probability 17 can not be updated as such by the combined segmentation and classification 100, but just added to the graph. The second character candidate is equal to the hypothesis ID 127a, in other words, the hypothesis ID is chosen as the second candidate character. The position of the second character candidate is provided by the segmentation part where the second character candidate has been identified, i.e., by the segmentation point i and the width j that have been considered. The probability of the second candidate character is equal to 1-ΡβΓΓ, with the PeiT value of 127b.
    The updating of graph 20 (Fig. 1) is preferably performed immediately after the selection of hypothesis ID 127 as the second character candidate by step 128. In other words, the update of graph 20 is preferably performed for each segmentation portion 125 that provides a hypothesis ID 127a that has a likelihood parameter 127b that satisfies the first predetermined criterion 128.
    The character classification 126 may provide a plurality of assumptions ID 127a each with their respective error likelihood Perr 127b, all corresponding to the same segmentation portion 125. In such a case, the checks 128 and 140 are performed on the hypothesis ID with the lowest Ρβίΐ, that is, the hypothesis ID that is expected to be the most correct. The "second character candidate with the position and probability" of the figures then comprises the plurality of ID hypotheses, each with their respective position and probability. They are all included in graph 20 directly before moving to a new segmentation segment 125.
    In combined segmentation and classification 100, any segmentation portion 125 is considered only once.
    The first predetermined criterion provides an adjustable parameter for determining the probability threshold from which the assumptions ID 127a are added in the graph. A very high Therr provides a very large second graph 21 which gives a high accuracy, but could slow down the OCR process 10. A lower Therr provides a second smaller graph 21, which could give a lower accuracy, but accelerates the OCR process 10.
    In one embodiment of the invention, different criteria could be used to determine whether the second character candidate, its position and its probability 17 are used in an update of the graph, if the update 110 of the stack is performed, if the update 150 of the width list is performed and / or if another width (j = j + 1) is considered.
    If the first predetermined criterion is satisfied, the combined segmentation-classification 100 checks in step 140 whether the error probability Perr satisfies a second predetermined criterion, which is that Perr is smaller than a threshold to have a low probability. Terr-Tisrr error is preferably chosen between 5% and 50%, for example, equal to 20%. In this way, the second predetermined criterion is satisfied only in cases where the hypothesis ID is really likely. If the second predetermined criterion is not satisfied, the index j is incremented and a next width from the stack of widths 131 is considered. If the second predetermined criterion is satisfied, the combined segmentation-classification 100 checks in step 141 whether the stack 102 still contains at least one segmentation point.
    If the stack 102 still contains at least one segmentation point, the index i of segmentation points is incremented and a next segmentation point 121 is considered.
    If the stack 102 does not contain a segmentation point, the combined segmentation-classification 100 preferably performs an end-of-stacking step 142. The stacking end step 142 includes a check that the end of the image of the string 11 has been reached, for example by verifying that HD of the rightmost character in the second graph 21 comprises the rightmost black pixel of the image of the string 11. IF the end of the image of string 11 has not been reached, an empty character with a probability equal to 0 is inserted into the graph starting from the vertex corresponding to the rightmost segmentation point that has been considered, a segmentation point corresponding to the end of this empty character is placed in the stack 102 and the combined segmentation and classification 100 resume with this segmentation point.
    If the end of the image of the character string 11 has been reached, an empty character with a probability equal to 0 can be added to the second graph 21 between each vertex that is not connected to a next vertex (each point of segmentation for which no second character candidate 17 has been selected) and the next vertex.
    Fig. 5 shows the update 110 of the stack according to one embodiment of the invention. When it is performed, the update 110 of the stack takes as input the segmentation part 125 considered at this stage in the combined segmentation and classification 100 and the image of the string of characters 11. The update 110 the stack comprises a determination 111 of another segmentation point. This other segmentation point 112 is said to correspond to the beginning of a character to the right of the character that has just been identified in the segmentation portion 125 by the character classification 126. The other segmentation point 112 is preferably determined as the leftmost black pixel to the right of the segmentation portion 125, the space between characters may have a predetermined value or be calculated during the combined segmentation and classification 100,
    Alternatively, the other segmentation point 112 can be calculated, for example, during character classification 126 and provides update 110 of the stack.
    In one embodiment of the invention, the determination 111 of another segmentation point uses information from the database 132.
    The update 110 of the stack then checks in step 113 whether the other segmentation point 112 is currently present in the stack 102 or in the list of segmentation points already processed. Otherwise, the other segmentation point 112 is placed in the stack 102.
    Since a character is supposed to start a bit after the preceding character, this update 110 of the stack generates segmentation points that are likely starting points for a next character. Verification in step 113 prevents a segmentation point from appearing twice in stack 102.
    Fig. 6 illustrates a very simple example of the combined segmentation and classification 100 according to the invention. Fig. 7 illustrates the second graph 21 at the end of the combined segmentation and classification 100 shown in FIG. 6.
    Fig. 6a shows an image of a string 11 with the word batch and in which the letters I and o are so close that it is difficult to determine whether it is i and o or b. The first OCR 12 provided the IDs b and t with their position and probability (P ~ 80% for b and R "95% for t). From their position, the segmentation points 601 and 602 can be determined (Fig. 6a). The graph which is the first graph at this stage comprises the vertices 711, 712 and 713 and the edges 701 and 702 (Fig. 7). The stack contains the segmentation points 601 and 602. For the segmentation point 601, the determination 130 of the width list determines possible widths including a width 603. The width 603 is placed in the list 131 of widths.
    Fig. 6b shows the width 603 which is considered the first along with the segmentation point 601 in the classification 126. The classification 126 finds that the segmentation part provided by the segmentation point 601 and the width 603 represents a b with a probability of 70%. Since Pei-r (30%) is less than Therr (99.9%), a corresponding edge 703 is added to graph 21 during an update of the graph. Update 110 of the stack determines the other segmentation point (step 111). It is found in step 102 that this other segmentation point that corresponds to point 602 is already in stack 102. Consequently, this other segmentation point is not added to stack 102. Update 150 the average width is performed, which updates the database 132. The determination 130 of the width list is also made to take into account the updating of the database. Since Perr (30%) is not less than (20%), another width 604 of list 131 is tested.
    Fig. 6c shows the width 604 which is considered together with the segmentation point 601 in the classification 126. The classification 126 finds that the segmentation portion provided by the segmentation point 601 and the width 604 represents an I with a probability of 85%. Since Perr (15%) is less than Therr (99.9%), a corresponding edge 704 is added to graph 21 during an update of the graph and a vertex 714, which corresponds to a segmentation point 605. at the end of I, is added to graph 21.
    Another segmentation point 605 is then determined in step 111 of update 110 of the stack. Since point 605 is not in the stack (step 113), it is added to the stack. Update 150 of the average width and determination 130 of the width list are also performed.
    Since ΡβίΤ (15%) is less than Tlerr (20%), there is a check (step 141) to check whether there is any segmentation point left in the stack 102, segmentation points 602 and 605 still being in the stack 102, the next segmentation point 602 is considered,
    Fig. 6d shows the segmentation point 602 considered together with the width 606 in the classification 126. The classification 126 finds that the segmentation portion provided by the segmentation point 602 and the width 606 represents a t with a probability of 99.5%. Since Perr (0.5%) is less than Therr (99.9%), a corresponding edge 705 is added to graph 21 during an update of the graph. Update 110 of the stack is not performed because the end of the string has been reached. The update 150 of the average width and the determination 130 of the width list are made,
    Since Pm (0.5%) is less than Tierr (20%), there is a check (step 141) to check whether there is any segmentation point left in the stack 102. The segmentation 605 is still in the stack 102, this segmentation point 605 is considered.
    Fig. 6e shows the segmentation point 60S considered together with a width 607 in the classification 126. The classification 126 finds that the segmentation portion provided by the segmentation point 605 and the width 607 represents an o with a probability of 85%. Since Ρθΐτ (15%) is smaller than Tl "w (99.9%), a corresponding edge 706 is added to graph 21 during an update of the graph The update 110 of the stack determines the other segmentation point (step 111) It is found in step 112 that this other segmentation point corresponding to point 602 is in the list of segmentation points that have already been processed. Segment 150 is not added to the stack 102. The update 150 of the average width and the determination 130 of the width list are also performed.
    Since Perr (15%) is less than Tlerr (20%), there is a check (step 141) to check if there is any segmentation point left in the stack 102. As the stack 102 is empty the combined segmentation and classification 100 moves to the end of stacking steps 142.
    The end of stacking steps verify that the rightmost pixel of the image of the character string 11 is actually a part of the t. All vertices of the graph 21 are connected to a next vertex and, consequently, no empty character is added to the graph 21.
    Decision models can then be added to the second graph shown in FIG. 7. Decision models can modify the probabilities shown in the graph. A contextual decision model based on an English dictionary can, for example, strongly favor the "batch" character string with respect to the "bt" string.
    The present invention comprises a combination of a first OCR 12 and a second OCR, which is the combined segmentation and classification 100, at various stages: at the input of the first segmentation points 14 of the stack 102 of the second OCR 100, the update 20 of the graph generated by the first OCR 12 which generates the second graph 21 and in the decision models 22. The invention does not require to have an overview of how the first OCR 12 is performed. It only requires obtaining from the first OCR 12 the character ID sequence, their position in the string image and, if possible, their likelihood parameters, which are the usual outputs of any As a result, the invention is capable of combining almost any commercial OCR with the second OCR.
    Furthermore, the invention takes the best of the two OCRs so that the output of the OCR method 10 according to the invention is better than the output of the first OCR 12 and better than the output of the second OCR 100, IF, moreover the first OCR 12 is very fast, the calculation time taken by the OCR method 10 according to the invention is not much higher than the calculation time taken by the second OCR 100 followed by decision models.
    All inclusive, in the OCR method according to the invention, two segmentations are performed. A first segmentation is performed during the first OCR 12, a second segmentation based on segmentation points and widths and which is combined with the classification as described with reference to FIG. 4, is performed during the second OCR 100. The second segmentation uses the segmentation points found by the first segmentation as input.
    According to one embodiment of the invention, at least one additional OCR is performed after the second OCR. The at least one additional OCR uses the segmentation points provided by one or more previously performed OCRs and provides character IDs with their positions and likelihood parameters that are used to update the graph prior to the addition of decision models. 22.
    According to one embodiment of the invention, at least part of the method according to the invention can be executed by a computer.
    In other words, the invention relates to a method and a program for combining a first optical character recognition (OCR) 12 and a second OCR 100. The first OCR 12 operates first on an image of a string of characters 11. Its output 13 (first characters identified, character positions and character likelihood parameters) is used to generate a first graph 16. Segmentation points related to the positions of the first identified characters 14 are used as entered by the second OCR 100 by performing a combined segmentation and classification on the string image 11. The output 17 (second identified characters, character positions and character likelihood parameters) of the second OCR 100 is used to set day 20 the first graph 16, generate a second graph 21 which combines the output 13 of the first OCR 12 with the output 17 of the Second OCR 100. Decision models are then used to modify the weightings 22 of the paths in the second graph 21 to generate a third graph 23. A better path 24 is determined on the third graph 23 to obtain the identification of the characters. present in the image of ka string 11.
    权利要求:
    Claims (24)
    [1]
    claims
    A method of selecting candidate characters in a method of identifying characters in a digital image, the method comprising the steps of: a) applying a first character identification process to determine first character candidates and a list of characters; segmentation points of the first character candidates, b) generating a list of character widths corresponding to a segmentation point of between the segmentation point list, c) determining a portion of the digital image corresponding to the segmentation point and a character width of the character width list; d) applying a character classification method on the digital image portion to obtain a hypothesis ID of a character possibly present in the portion of the digital image and a likelihood parameter which relates to the likelihood that the ID hypothesis is correct and e) selecting the hypothesis ID as the second character candidate in the numerical image if the likelihood parameter fulfills a first predetermined criterion,
    [2]
    The method of claim 1, the method further comprising a step f) updating the list of segmentation points with another segmentation point determined on the basis of the portion of the digital image corresponding to the second character selected candidate,
    [3]
    The method of claim 2, wherein step f) comprises: checking whether the other segmentation point is already in the segmentation point list and the addition of the other segmentation point to the list of segmentation points if the other segmentation point is not already present in the segmentation point list,
    [4]
    The method of claim 1, wherein in step b) the character width list is generated based on at least the list of segmentation points of the first character candidates.
    [5]
    The method of claim 1, wherein steps b, c, d and e are performed for each segmentation point of the segmentation point list,
    [6]
    The method of claim 1, wherein steps c, d and e are repeated for another character width of the character width list if the likelihood parameter does not satisfy the first predetermined criterion,
    [7]
    The method of claim 1, wherein steps c, d and e are repeated for another character width of the character width list if the likelihood parameter satisfies the first predetermined criterion and does not satisfy a second criterion. predetermined.
    [8]
    The method of claim 1, wherein the method further comprises the steps of: f) generating a data structure from the first character candidates and their segmentation points and g) updating the data structure with the second candidate character selected and a next segmentation point calculated from the character width of the second selected character candidate and the segmentation point used to determine the digital image portion corresponding to the second selected character candidate,
    [9]
    The method of claim 8, wherein step a) comprises determining first likelihood parameters of the first candidate characters providing an indication of the chance that the candidate character is correct, the data structure comprises the first likelihood parameters. first candidate characters and the method further comprises the steps of: changing the scale of the first likelihood parameters of the first candidate characters and / or the likelihood parameters of the second candidate characters to make them comparable to each other and updating the data structure with the likelihood parameters of the second candidate characters.
    [10]
    The method of claim 9, wherein the data structure is a graph in which the segmentation points are represented by vertices and the candidate characters are represented by edges between the vertices.
    [11]
    The method of claim 10, further comprising the step of applying a decision modeling method to the updated data structure.
    [12]
    The method of claim 11, further comprising the steps of: determining a better path in the updated data structure and selecting the candidate characters corresponding to the edges of the best path as the digital image characters.
    [13]
    The method according to claim 12, wherein the step of applying a decision modeling method comprises advancing the candidate characters that have the first likelihood parameters of the first character candidates and the second likelihood parameters. second candidate characters satisfying a third predetermined criterion.
    [14]
    The method according to claim 13, wherein some of the first and second character candidates are characters of a first or second type, and the decision modeling method includes the advantage of the first character type candidates. whether the candidate characters are provided by the first character identification process and favoring the second type of candidate candidates if the candidate characters are provided by the classification process.
    [15]
    The method of claim 1, wherein the first character identification method comprises a decision modeling method,
    [16]
    The method of claim 1, wherein the first character identification process provides positions of first character candidates and further comprises a step of determining the list of segmentation points of the first character candidates from the positions of the first characters. candidate characters.
    [17]
    A method of identifying characters in a digital image, the method comprising the steps of: i) applying a first character identification method for determining first character candidates and first segmentation points corresponding to the first character candidates, b) generating a first graph on the basis of the first character candidates and the first segmentation points, in which the first segmentation points are represented by vertices and the first candidate characters by edges, c) applying a second character identification method on at least a portion of the digital image to generate at least a second character candidate, d) updating the graph by adding the at least one second character candidate to the first graph and e) selecting in the updated graph candidate characters such as characters of the digital image.
    [18]
    The method of claim 17, wherein the second character identification method generates at least one second segmentation point and wherein the at least one second segmentation point is added to the graph in step d).
    [19]
    The method of claim 18, wherein the first character identification method provides first likelihood parameters providing a likelihood that the first character candidates are correct, the second character identification method provides second likelihood parameters providing a likelihood that the at least one second character candidate is correct, and the method further comprises the steps of: changing the scale of the first likelihood parameters and / or the second likelihood parameters to make them comparable to each other; other, add the first likelihood parameters in the first graph and add the second likelihood parameters in the updated graph.
    [20]
    The method of claim 19, further comprising the step of applying a decision modeling method to the updated graph.
    [21]
    The method of claim 20, wherein the decision modeling method comprises a rule that favors the candidate characters that have the first likelihood parameter and the second likelihood parameters that satisfy a third predetermined criterion.
    [22]
    The method of claim 21, wherein the first character identification method provides candidate characters for characters of a first type, wherein the second character identification method provides candidate characters for characters of a second type and wherein the decision modeling method comprises a rule which favors the candidate characters corresponding to the first type of characters if these candidate characters have been determined by the first character identification method and favors the candidate characters corresponding to the second character type; type of characters if these candidate characters were determined by the second character identification method,
    [23]
    The method of claim 17, wherein the second character identification method uses as input, in step c), the first segmentation points from the first character identification method.
    [24]
    A computer-readable non-transitory medium storing a program causing a computer to execute a method for identifying characters in a digital image, the method comprising the steps of: a) applying a first character identification method to determine first characters; candidate characters and a list of segmentation points of the first candidate characters, b) generate a list of widths of characters corresponding to a segmentation point of the list of segmentation points, c) determine a portion of the digital image corresponding to the segmentation point provided by the position of a first candidate initial character and a character width from the list of character widths, d) applying a character classification method on the digital image portion to obtain a hypothesis ID of a character possibly present in the part of the digital image and a likelihood parameter which relates to the likelihood that the hypothesis ID is correct and e) selecting the hypothesis ID as a character in the digital image if the likelihood parameter satisfies a first predetermined criterion.
    类似技术:
    公开号 | 公开日 | 专利标题
    BE1024194A9|2017-12-19|Method for identifying a character in a digital image
    BE1022562B1|2016-06-02|Optical character recognition method
    EP1836651B1|2018-02-21|Method for searching, recognizing and locating a term in ink, and a corresponding device and computer program
    FR2982388A1|2013-05-10|SYSTEM AND METHOD FOR PRODUCTIVE GENERATION OF COMPOUND WORDS IN AUTOMATIC STATISTIC TRANSLATION
    Bluche et al.2014|The a2ia arabic handwritten text recognition system at the open hart2013 evaluation
    BE1025503B1|2019-03-27|LINE SEGMENTATION METHOD
    FR2963695A1|2012-02-10|POLICE WEIGHT LEARNING FOR TEST SAMPLES IN MANUSCRIPTED KEYWORD REPORTING
    US20190130232A1|2019-05-02|Font identification from imagery
    FR2911201A1|2008-07-11|Written text editing method for correcting spelling error, involves calculating difference between apparition frequency of one n-gram in text and in language using n-gram by n-gram technique
    KR20200055760A|2020-05-21|Image content recognition method and device
    EP1669886A1|2006-06-14|Construction of an automaton compiling grapheme/phoneme transcription rules for a phonetiser
    JP2020095713A|2020-06-18|Method and system for information extraction from document images using conversational interface and database querying
    BE1025504A1|2019-03-20|Pattern recognition system
    CN109948615B|2021-01-26|Multi-language text detection and recognition system
    BE1025502A1|2019-03-20|SYSTEM AND METHOD FOR FORM RECOGNITION USING GABOR FUNCTIONS
    BE1025134B1|2018-11-16|Method of identifying a character in a digital image
    BE1026039B1|2019-12-13|IMAGE PROCESSING METHOD AND IMAGE PROCESSING SYSTEM
    BE1025006B1|2018-09-25|COMPUTER-IMPLEMENTED PROCESS AND OPTICAL CHARACTER RECOGNITION SYSTEM
    FR2880708A1|2006-07-14|Term e.g. typed character, searching method for digital handwritten document, involves converting handwritten data into intermediate data, in intermediate format, in form of segmentation graph, and searching terms on intermediate data
    Idziak et al.2021|Scalable Handwritten Text Recognition System for Lexicographic Sources of Under-Resourced Languages and Alphabets
    BE1020588A5|2014-01-07|METHOD OF RECOGNIZING FORMS, COMPUTER PROGRAM PRODUCT, AND MOBILE TERMINAL.
    Slimane et al.2012|Benchmarking Strategy for Arabic Screen-Rendered Word Recognition
    Kurhekar et al.2021|Automated Text and Tabular Data Extraction from Scanned Document Images
    CN113378562A|2021-09-10|Word segmentation processing method and device, computing equipment and storage medium
    WO2019155168A1|2019-08-15|Method for transforming a sequence so as to render it executable by a machine
    同族专利:
    公开号 | 公开日
    BE1024194B1|2017-12-12|
    BE1024194A1|2017-12-06|
    WO2017064272A3|2017-05-26|
    US20170109573A1|2017-04-20|
    WO2017064272A2|2017-04-20|
    US9836646B2|2017-12-05|
    引用文献:
    公开号 | 申请日 | 公开日 | 申请人 | 专利标题
    US551978A|1895-12-24|Bath-brush apparatus |
    US5519786A|1994-08-09|1996-05-21|Trw Inc.|Method and apparatus for implementing a weighted voting scheme for multiple optical character recognition systems|
    US6504540B1|1995-06-19|2003-01-07|Canon Kabushiki Kaisha|Method and apparatus for altering one or more attributes of one or more blocks of image data in a document|
    WO1997035561A1|1996-03-28|1997-10-02|The Board Of Trustees Of The University Of Illinois|Materials and methods for making improved liposome compositions|
    US6473517B1|1999-09-15|2002-10-29|Siemens Corporate Research, Inc.|Character segmentation method for vehicle license plate recognition|
    WO2002037933A2|2000-11-08|2002-05-16|New York University|System, process and software arrangement for recognizing handwritten characters|
    WO2002089105A2|2001-05-02|2002-11-07|Bitstream, Inc.|Methods, systems, and programming for producing and displaying subpixel-optimized images and digital content including such images|
    FI120165B|2004-12-29|2009-07-15|Seven Networks Internat Oy|Synchronization of a database through a mobile network|
    US8644611B2|2009-06-03|2014-02-04|Raytheon Bbn Technologies Corp.|Segmental rescoring in text recognition|
    JP5591578B2|2010-04-19|2014-09-17|日本電産サンキョー株式会社|Character string recognition apparatus and character string recognition method|
    US9152871B2|2013-09-02|2015-10-06|Qualcomm Incorporated|Multiple hypothesis testing for word detection|
    US9183636B1|2014-04-16|2015-11-10|I.R.I.S.|Line segmentation method|US10191986B2|2014-08-11|2019-01-29|Microsoft Technology Licensing, Llc|Web resource compatibility with web applications|
    US9805483B2|2014-08-21|2017-10-31|Microsoft Technology Licensing, Llc|Enhanced recognition of charted data|
    US9524429B2|2014-08-21|2016-12-20|Microsoft Technology Licensing, Llc|Enhanced interpretation of character arrangements|
    US9397723B2|2014-08-26|2016-07-19|Microsoft Technology Licensing, Llc|Spread spectrum wireless over non-contiguous channels|
    US10878269B2|2018-06-19|2020-12-29|Sap Se|Data extraction using neural networks|
    US10824811B2|2018-08-01|2020-11-03|Sap Se|Machine learning data extraction algorithms|
    US10803242B2|2018-10-26|2020-10-13|International Business Machines Corporation|Correction of misspellings in QA system|
    CN112085019A|2020-08-31|2020-12-15|深圳思谋信息科技有限公司|Character recognition model generation system, method and device and computer equipment|
    法律状态:
    2018-02-15| FG| Patent granted|Effective date: 20171212 |
    优先权:
    申请号 | 申请日 | 专利标题
    US14884361|2015-10-15|
    US14/884,361|US9836646B2|2015-10-15|2015-10-15|Method for identifying a character in a digital image|
    [返回顶部]