• 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:
    Group images into at least two clusters based on color content similarity, identify at least one representative image in each cluster, and all other images within each cluster are identified as non-representative images, e.g., loss or Using a lossless coding algorithm, independently coding a representative image from each cluster, and predictively coding each non- representative image from each cluster using the representative image as a reference image from the cluster. Way.
    公开号:KR20010025113A
    申请号:KR1020007013547
    申请日:2000-03-09
    公开日:2001-03-26
    发明作者:크리쉬나마카리산타나
    申请人:요트.게.아. 롤페즈；코닌클리케 필립스 일렉트로닉스 엔.브이.；
    IPC主号:
    专利说明:

    Image compression
    The term "image" throughout this specification refers to a still picture as opposed to a moving picture. Moreover, in general, it can be appreciated by those skilled in the relevant art that the image compression techniques described herein can be used to compress images contained in an image database for transmission, processing, editing, manipulation, or any other purpose. Will be recognized. In this regard, use of the present application is not limited to any particular application, and the present invention is not limited to any particular application.
    Typically, when a set of images is compressed, each individual image is compressed independently using either a lossless or lossy compression scheme. Image compression mechanisms, such as those specified by the JPEG standard, use spatial correlation to represent an image with fewer bits to achieve compression.
    The present invention generally relates to image compression.
    1 is a diagram illustrating the concept of a hierarchical clustering algorithm used in connection with the image compression technique of a sample embodiment of the present invention.
    2 is a diagram illustrating the concept of predictive coding of a target image from a reference image in accordance with the presently preferred embodiment of the image compression technique of the present invention.
    3 shows an embodiment of a transmitter according to the invention.
    4 shows an embodiment of a storage device according to the invention.
    5 shows an embodiment of a display device according to the invention.
    It is an object of the present invention to provide better compression of image sets. To this end, the present invention provides a method and apparatus for compression, a transmitter, a coded data stream, a storage medium, a method and apparatus and a device for decompression as defined in the independent claims of the claims. Advantageous embodiments are defined in the dependent claims of the claims.
    The invention, in its broadest aspect, designates at least one representative image or sub-image in an image or sub-image set, wherein all other images or sub-images in the image or sub-image set are non-representative images or sub-images. Said designation step; Independently coding the representative images or sub-images; And predictively coding each of said non-representative images or sub-images using said representative images or sub-images as reference images or sub-images.
    To date, no image compression technique is known which predictively codes one image against another to take advantage of redundancy between the images, thereby attempting to achieve a higher level of image compression. Motion coding is generally used to compress moving images by predictively coding some images with reference to other images to achieve higher levels of compression, although similar techniques have not been applied to image compression.
    Based on the above, in order to achieve a significantly higher level of image compression, image compression attempts to exploit not only the spatial correlation within a given image, but also the correlation between different images. It will be appreciated that there is a need for technology now. The present invention satisfies this need in the art by utilizing content-based similarity (used to search and search for similar images in an image database) that efficiently compresses a given set of images.
    The present invention is directed to grouping images into at least two clusters based on image similarity according to at least one image similarity measure, and identifying at least one representative image in each cluster, wherein All the other images in the cluster are identified as non-representative images, and independently coding a representative image from each cluster using, for example, a lossy (eg JPEG) or lossless coding algorithm; And predictively coding each non-representative image from each said cluster using the representative image (s) from that cluster as a reference image. In the presently preferred embodiment, the image similarity measure comprises a color similarity measure. However, the image similarity measure can be a suitable image similarity measure or a combination thereof, eg, color similarity measure, texture similarity measure, shape similarity measure, geometrical similarity measure, and / or content similarity measure.
    The invention further comprises subdividing each image into two or more partitions, based on image similarity according to at least one similarity measure of the corresponding partitions of the partitions having corresponding positions within each image. Grouping into at least two clusters, thereby generating a set of clusters for each corresponding set of ones of the compartments, and grouping the at least one representative image for each partition cluster Identifying, wherein every other image in each cluster is identified as a non-representative image, independently coding a reference partition in a representative image for each cluster, and representative for a cluster that is a reference partition. For each of these clusters using a reference section within the image. It includes a respective set of image compression methods, comprising the step of predicting the coded blocks in the non-representative images.
    In one embodiment, the method further comprises subdividing each partition of each non-representative image into a number of smaller target blocks for each cluster, dividing each partition of the representative image (s) with the target blocks. Subdividing into a plurality of smaller reference blocks having a size, wherein the predictive coding step comprises:
    For each target block, comparing the pixels of the target block with the pixels of each equally sized reference blocks in the reference partitions according to the exact search metric, and producing an error metric value for each comparison,
    For each target block, determining if any error metric value produced is less than a prescribed maximum threshold, and
    If so, identifying one of the equally sized reference blocks making up the best match for the target block, and
    If not, it is performed for each cluster by independently coding the corresponding target block.
    In another embodiment, the predictive coding step is performed for each cluster by coding the difference between each non-representative partition and the reference partition for that cluster.
    The present invention also provides for specifying at least one representative sub-image in the set of sub-images, wherein all other sub-images in the set of sub-images are designated as non-representative sub-images; Independently coding the representative sub image (s); And predictively coding each of said non-representative sub-images using a representative sub-image as reference sub-image (s).
    The invention also includes subdividing each image into two or more compartments; Grouping the partitions into at least two clusters based on image similarity according to at least one similarity measure; Identifying at least one representative compartment for each cluster of compartments, wherein all other compartments in each of the clusters are identified as non-representative compartments; Independently coding the representative partitions for each of the clusters; And predictively coding respective non-representative partitions in each of said clusters using a representative partition for the cluster as a reference partition. In one embodiment, the method further comprises subdividing each non-representative partition into a plurality of smaller reference blocks having the same size as the plurality of smaller target blocks for each cluster, wherein the predictive coding The steps are,
    For each target block, the pixels of that target block are compared with the pixels of each of the same sized reference blocks in the reference partition (s) according to the defined search metric, and the error metric value for each comparison. Producing step,
    For each target block, determining if any error metric value produced is less than an accurate maximum threshold, and
    If so, identifying one of the equally sized reference blocks making up the best pair for the target block, and
    If not, it is performed for each cluster by independently coding the corresponding target block.
    In another embodiment, the predictive coding step is performed by coding the difference between each non-representative partition and the reference partitions for that cluster for each cluster.
    According to any other feature of the invention, if a particular target image or section is larger than one reference image or section in the cluster, one of the most similar or at least the most similar reference images or sections, eg, One of the reference images or partitions sharing the closest common ancestor node in the cluster tree for that node is selected as the reference image or partition for the particular target image or partition.
    In a presently preferred embodiment, the method
    Preliminary partitioning each image into a plurality of color comparison blocks; A preliminary step of calculating a normalized histogram for each of said color comparison blocks of each said image; And performing a preliminary step of using the calculated and standardized histogram in the step of performing the grouping and identification step.
    In a sample embodiment, each non-representative image or sub-image comprises partitioning the non-representative image or sub-image into a plurality of target blocks of pixels; For each of the target blocks, the pixels of that target block are compared with the pixels of each of the same sized reference blocks in a reference image or sub-image according to a defined search metric, and an error for each comparison. Producing a metric value; For each target block, determining if any of the produced error metric values is less than a prescribed maximum threshold, and
    If so, identifying one of said equally sized reference blocks making up the best pair for that target block, and
    If not, then it is coded for the selected reference image or sub-image by the step of coding the target block independently.
    The coded data stream identifies the reference blocks of the reference images that constitute the best pair for each target block of each of the non-representative images that are not independently coded, and the target block of each non-representative image. And may further comprise coded data identifying whether they are coded independently, and furthermore, identifying whether the images of the set of images are reference images or non-representative images. Alternatively, or additionally, the predictive coding step may be performed for each target block that is not coded independently, to produce a residual value for each of the target blocks that are not coded independently, and the target block and its target block. And calculating a difference between the reference blocks forming the best pair, and coded data representing the calculated residuals can also be injected into the coded data stream.
    The invention also includes an apparatus for implementing the methodology of the invention.
    These and other features, forms, and advantages of the present invention will be readily understood from the following detailed description taken in conjunction with the accompanying drawings.
    A hierarchical clustering algorithm for grouping images based on color similarity is described in unpublished US patent application Ser. No. 09 / 102,474, corresponding to PCT / IB99 / 01008 (our reference PHN 15.961). A constant image set as input, the hierarchical clustering algorithm generates a cluster set as output. The images in each cluster are similar to each other in terms of the spatial distribution of color. In the above-referenced patent application, the technique is used in the context of an image retrieval system for retrieving and retrieving similar images in an image database for retrieving images. The hierarchical clustering algorithm also provides for selecting a representative image from each cluster.
    In general, the hierarchical clustering algorithm described above works by partitioning each image into a number of color comparison partitions, and then calculates a normalized histogram for each color comparison partition of each image. The normalized histogram thus calculated is then used to group the images into optimal clusters, and then to select one or more representative images for each cluster. The details of this hierarchical clustering algorithm used in connection with the presently preferred embodiment of the present invention can be explained by reference to the above-mentioned patent application. However, it should be clearly understood that the present invention is not limited to any particular image grouping or clustering algorithm, or to the use of any image grouping or clustering scheme, in its broadest sense.
    FIG. 1 shows how the hierarchical clustering technique described in the application PCT / IB99 / 01008 referenced above works with an exemplary case of eight image sets numbered from 1 to 8. FIG. First, color histograms are calculated for each image. Second, the similarity of all images is calculated using color histograms. Third, the images with the closest similarity are grouped. For example, images 2 and 4 are grouped in the form of node 10 in the cluster tree. This process is repeated until the root node of the cluster tree arrives. At the end of the hierarchical clustering process, in the example case shown in FIG. 1, two groups of eight images, one group with five images (image numbered 1-5), and three images Clustered into different groups (images numbered 6-8). For the first cluster, the representative images are 2 and 4, and for the second cluster, the representative image is 7. The representative image is shown in the shaded portion in FIG. 1. Because of the nature of the clustering process, each cluster has an associated cluster tree. Unlike the above example case, the number of representative images is about 10-20% of the total number of images in the cluster.
    From the clustering theory, it can be inferred that the images in the cluster are similar to each other. Therefore, instead of attempting to compress these images independently, it is wise to use the similarity between the images to efficiently improve image compression. More specifically, the image compression mechanism according to the presently preferred embodiment of the present invention includes the following steps.
    1. grouping a given set of images into clusters based on color content similarity and identifying / designating at least one representative image for each cluster, wherein all remaining images in each cluster are identified / designated, The identification / designation step.
    2. Compressing representative images in each cluster using either high performance lossless image coding technique or lossy coding technique. This step consists of, for example, a JPEG coding scheme for the case of lossy coding. However, the particular coding mechanism used does not limit the invention in its broadest sense.
    3. In each cluster, coding the rest of the images using the representative image as the reference image. In order to code a given non-representative image (hereinafter referred to as a "target image"), the target image is first decomposed into many target blocks. For example, the target blocks may be 8x8 or 16x16 pixel blocks. For each target block in the target image, a similar block is searched for in the reference image. Once such a reference block is found, the difference between the reference block in the reference image and the target block in the target image is coded. The remainder (difference between the target block and the reference block) is coded using known discrete cosine transform (DCT) and variable length coding (VLC) coding techniques. For a given target block, if there are no similar reference blocks provided in the reference image (s), the target block is coded unpredictably (ie, independently-without reference to any reference image).
    In general, as readily understood by those skilled in the relevant art, each target block in a target image is best matched by a search metric (i.e., matching standard) in which equally sized reference blocks within a specified horizontal and vertical search range are specified. It may be compared with all possible equally sized reference blocks of the reference image located within the specified horizontal and vertical search ranges of the corresponding position of the current target block within the reference image to determine whether to construct a pair. The hardware and / or software that performs the search is generally referred to as a "search engine." And there are many known standards for determining the quality of a partner.
    The most well known standard (i.e., search metric) is the minimum absolute error, where the search metric consists of the sum of the absolute values of the difference between each corresponding pixel in the reference block and each pixel in the target block ( MAE), and between the Minimum Square Error (MSE), where the search metric consists of the sum of the squares of the pixel differences above, in either case, the pair with the smallest value of the corresponding sum is The best pair is chosen within the specified search range, and therefore the horizontal and vertical position relative to the current target block is determined by the best pair vector that can be coded using DCT or other suitable coding technique to achieve better image compression. Configure. If the resulting minimum sum (i.e. error metric) is nevertheless considered too large, the decision is that no suitable pair exists for the current target block, in which case the target block is independently coded into any reference block in the reference image. It can be made to be. For the purposes of the present invention, one of the above two standards, or any other suitable standard, may be used.
    With continued reference to FIG. 1, the image compression mechanism according to the first optional embodiment of the present invention includes the following steps.
    1. Subdivide each image in a given set of images into two or more partitions (or “sub-images”), eg, a 256x256 pixel image can be subdivided into sixteen 64x64 pixel partitions. step.
    2. For example, a set of 16 different clusters with an image subdivided into 16 different partitions is based on the obtained color content similarity (eg, based on the similarity of the color histograms of the partitions). Grouping the compartments into other clusters. For example, if the partitions are numbered from 1 to 16, starting from the upper left corner and ending with the lower right corner, all the partitions numbered as 1 are grouped into one or more clusters, and 2 In general, all partitions numbered n (hereinafter referred to as partition “n”, 1 to 16) are grouped into clusters, such that all numbered partitions are grouped into one or more clusters, and each The set of clusters of results in n sets of total clusters containing two or more clusters.
    3. Identifying / designating at least one representative image for each cluster partitions, wherein the remaining image of each cluster is designated as a non-representative image.
    4. Subdividing each section of each image into a number of smaller blocks (eg, 8x8 or 16x16 pixel blocks) to facilitate the predictive coding process. In each cluster partition, the reference partitions in the representative image for the cluster are independently clustered (compressed) using either a lossless image coding technique or a lossy coding technique with very high performance. This can be done with a JPEG coding scheme for the case of lossy coding, for example. However, the specific coding mechanism used is, in the broadest sense, not limited to the present invention.
    5. In each cluster partition, for each target block in the non-representative image, a similar block is searched in the corresponding (reference) partition (s) of the representative image (s) for the cluster. Once such a reference block is found, the difference between the reference block in the reference partition and the target block in the target partition is coded. The remainder (difference between the target block and the reference block) is coded using known discrete cosine transform (DCT) and variable length coding (VLC) coding techniques. For a given target block, if there are no dissimilar reference blocks provided within the corresponding reference section of the reference image, the target block is coded unpredictably (ie, independently-without reference to any reference section). For example, referring to FIG. 1, an image 5 has been selected as a representative image for the cluster containing compartment 1 of image 2, and image 4 is a cluster containing compartment 2 of image 2. Assuming that it is selected as the representative image for, for the target block in the partition 1 of the image 2, the reference block is selected from the partition 1 of the representative image 5 and the partition 2 of the image 2. For the target block in), the reference block is selected from the section 2 of the representative image 4.
    In general, as can be readily appreciated by those skilled in the relevant art, each target block within a target partition is assigned a search metric (i.e., a matching standard) with equally sized reference blocks within a specified horizontal and vertical search range. It can be compared with all possible equally sized reference blocks of the reference partition located within the specified horizontal and vertical search ranges of the corresponding positions of the current target block within the reference block to determine if they constitute the best pair according.
    With continued reference to FIG. 1, an image compression mechanism according to a second optional embodiment of the present invention includes the following steps. Subdividing each image in a set of images into two or more partitions (or “sub images”), such as, for example, 256x256 pixel images can be subdivided into sixteen 64x64 pixel partitions.
    2. Group all the partitions into different clusters based on color content similarity (eg, based on the similarity of the color histogram of the partition) without reference to a position in each image. Thus, as with the first optional embodiment, only the other cluster's partitions will be obtained, against the other set of clusters for each set of corresponding partitions.
    3. Identifying / designating all remaining compartments in each cluster compartment designated as at least one representative compartment, non-representative compartment for each cluster compartment.
    4. Subdividing each section of each image into a number of smaller blocks (eg, 8x8 or 16x16 pixel blocks) to facilitate the predictive coding process. In each cluster partition, the reference partition for the cluster is independently coded (compressed) using either a lossless image coding technique or a lossy coding technique with very high performance. This can be done, for example, with a JPEG coding scheme for the case of lossy coding. However, the specific coding scheme used does not limit the invention in its broadest sense.
    In each cluster partition, for each target block in the non-representative partition, a similar block is searched within the reference partition for the cluster. Once such a block is found, the difference between the reference block in the reference partition and the target block in the target partition is coded. The remainder (difference between the target block and the reference block) is coded using known discrete cosine transform (DCT) and variable length coding (VLC) coding techniques. In certain target blocks, if there is no similar reference block provided within the reference partition, the target block is coded unpredictably (ie, independently-without reference to any reference partition).
    According to another alternative embodiment of the present invention, step 4 (ie, subdividing each partition into a plurality of blocks) of each of the first and second alternative embodiments described above may be deleted. Accordingly, step 5 (ie, predictive coding step) may be performed by directly coding the difference between the target and the reference partition, rather than searching for the best pair on the block x block basis. Of course, this technique will require better partitioning of the image within the clustering process that will involve additional computational overhead at the front end to achieve a constant image compression quality. However, by eliminating the need for block x blocks to search for the best pair at the back end, the computational overhead associated with the process is eliminated. Balance within computational overhead, image compression quality, cost, complexity, etc. will vary depending on the size of the database, such as the similarity and shape of the image, due to the particular application, for example, the image database.
    With respect to any of the embodiments of the present invention described above, when a particular ticket image or partition has one or more reference images or partitions within its cluster, a single of the reference image or partition may be in any suitable way the specific target image or partition. Can be selected for. For example, referring to FIG. 1, an advantage due to the nature of the clustering process can be obtained that each cluster will have a cluster tree. As such, it provides a convenient way of identifying whether a reference image or partition in the cluster is nearly similar or statistically almost similar (ie, almost identical) to the particular non-representative image or partition.
    To illustrate this point in more detail, consider the case of the coded image 4 of FIG. 1. Thus, the "best pair" reference block will be found for each target block in the image 4. From the cluster tree for the cluster to which an image 4 belongs, the common that the images 2 and 4 share (which is the exact reason why the images 2 and 4 were clustered together before the image 5 joined the cluster). It can be inferred that they are more similar to images 4 and 5 because they share a common ancestor node (node 10) that is closer than the ancestor node (node 14). Thus, according to this additional optional feature of the invention, the image 2 can be selected as a single reference image for the image 4 during the predictive coding process. In general, according to this additional feature of the present invention, if a particular target image or partition has more than one reference image or partition in the cluster, one that shares the closest common ancestor node in the cluster tree for the node. The reference image or section is selected as the reference image or section for the particular target image or section.
    2 shows that the first target block TB1 in the target image 4 is predictively coded from the reference image 2 using the reference block RB2, and the second target block TB2 in the target image 4 is referenced. The presently preferred embodiment of the image compression of the present invention, which is predictively coded using the reference block RB5 from the image 5, is schematically illustrated.
    Thus, the image compression / coding scheme of the present invention not only uses spatial redundancy in a constant image by using DCT coding technique, but also groups the images into clusters based on color similarity, and employs a predictive coding technique for further compressing image data. In some cases, it also takes advantage of redundancy between images. Since this mechanism calculates the similarity between images, it will provide a high compression ratio for the required SNR or visual quality.
    The procedure for a decoded image coded according to the above-described image compression method of the presently preferred embodiment of the present invention is different from a reference and an unreferenced image. The reference image is directly decoded by performing variable length decoding and inverse DCT of the individual blocks of the reference image. An unreferenced image is decoded in two steps. In order to decode each block in an unreferenced image, the reference block corresponding to one image of the reference image is first decoded by decoding the remaining values in the target block. The value of the reference block and the decoded remaining value are added to give the actual value of the target block.
    3 shows an embodiment of a transmitter 10 according to the invention comprising means 11 for obtaining a set of still images 1,... 8, for example a camera or some receiving means. The transmitter 10 further comprises an apparatus 12 for compressing the set of still images 1,... 8. The apparatus comprises means 13 for identifying a representative image or image segment in the manner described above. Preferably, the identification means 13 group the set of images in the cluster, while identifying the representative image for each cluster. The representative image is coded independently in the coding means 15. Other images are identified as non-representative images. The non-representative image is predictively coded in the coding means 14 having the reference image as the reference image. The reference image may also be a reconstruction of the independently coded representative image rather than the decoded representative image. Moreover, predictive coding may be performed in another domain (eg, frequency domain) using a representative image coded independently as a reference image. Referring to the numbers used in the example of FIG. 1, the set of still images 1,..., 8 is compressed by the device. In the identification means 13, the images are grouped in a cluster, and the images 2, 5 and 7 are identified as representative images. These images 2, 5, 7 are independently coded. The other images 1, 3, 4, 6, 8 are identified as non-representative images and predictively coded. Images 1, 3, 4 are thus encoded with images 2 and 5 as reference images, and images 6 and 8 are encoded by using image 7 as reference images. As a result, the compressed representative images (2 * , 5 * , 7 * ) and the compressed non-representative images (1 * , 3 * , 4 * , 6 * , 8 * ) are coded data streams ({1 * , .. , 8 * }) is multiplied within the multiplexer 16 which constitutes a set of compressed images 1 * ,..., 8 * .
    FIG. 4 shows an embodiment 20 similar to FIG. 3 arranged for storing a coded data stream representing a set of compressed still images {1 * ,..., 8 * } on the storage medium 17. . The multiplexer 16 has been adapted to supply the appropriate format to the storage medium 17.
    FIG. 5 shows a compressed still image (1 * , ..... Transmitted in the coded data stream, for example transmitted by the transmitter 10 shown in FIG. 3 or obtained from the storage medium 17 shown in FIG. , 8 * ) shows an embodiment of a display device 30 according to the invention, comprising means 31 for receiving a set. The display device further comprises a device 32 for decompressing a set of compressed images 1 * ,..., 8 * , and a display unit 33. The decompression device 32 comprises means 34 for identifying representative images 2 * , 5 * , 7 * in a set of compressed still images 1 * ,..., 8 * . The representative image 2 * , 5 * , 7 * independently decodes the independent decoding means 36 to obtain the decompressed still images 2, 5, 7. Said well decoded representative image 2, 5, 7 is used as reference image in predictive decoding means 35 to obtain decompressed non-representative images 1, 3, 4, 6, 8. . The decompressed images 1,..., 8 are combined in the combining means 37 for display on the display unit 33. Still images 1,..., 8 may be displayed in the combined image or selection may be performed.
    While several preferred and optional embodiments of the invention have been mentioned in the foregoing detailed description, it is apparent to those skilled in the relevant art that various modifications and / or modifications of the basic inventive concepts described herein are set forth in the appended claims. It is obvious that it is still included within the scope of the invention.
    For example, the above-described embodiments of the present invention group images (or compartments / subimages) into clusters, measure similarity between different images, and represent representative and non-representative images (or compartments / subimages) within each cluster. Although each of the clustering algorithms using color similarity measures are used to identify, it is desirable that any suitable image or algorithm and / or image similarity grouping, or sub-images clustering are used selectively or additionally in any suitable combination. It will be appreciated by those skilled in the art.
    For example, the image or sub-image similarity measurement can be, for example, any combination of color similarity measurement, texture similarity measurement, shape similarity measurement, geometric similarity measurement, and / or content similarity measurement, All these measurements are known in the art. Moreover, although the above-described embodiments of the present invention each use clustering algorithms for grouping images or sub-images into different clusters and for selecting at least one representative image or sub-image within each cluster for predictive coding purposes, In the broadest sense it does not require any such grouping or clustering of images or sub-images. In various applications, the technique will improve image compression that is efficiently achieved.
    Rather, the present invention designates, in the broadest sense, at least one representative image or sub-image within an image or sub-image set, an image that is designated as a non-representative image or sub-image, or any other image or sub-image within a sub-image set, Independently coding a representative image or sub-image, and predictively coding each non-representative image or sub-image using the representative or sub-image as a reference or representative image. do. In this regard, how the images or sub-images are organized, grouped, arranged or clustered, and how the representative or sub-images are selected are not the limitations of the present invention in the broadest sense.
    In the claims, any reference signs placed between insertion phrases shall not be construed as limiting the claim. The word "comprises" does not exclude the presence of other elements or steps than those listed in the claims. The invention can be implemented by means of hardware comprising some specific components and by means of a suitably programmed computer. In the device claim enumerating several means, several such means are realized by hardware and the same item of hardware.
    权利要求:
    Claims (16)
    [1" claim-type="Currently amended] In the method (12) for compressing a set of still images (1, ..., 8),
    The method,
    Identifying (13) at least one representative image (2, 5, 7) in the set of still images (1, ..., 8), the other still images (1, 3, 4, 6, 8) in the set; ) Is identified as a non-representative image, said identification step 13,
    Independently coding (15) said at least one representative image (2, 5, 7), and
    Predictively coding (14) each non-representative image (1, 3, 4, 6, 8) using the at least one representative image (2, 5, 7) as a reference image.
    [2" claim-type="Currently amended] The method of claim 1,
    Grouping the still images 1,..., 8 into at least two clusters based on image similarity according to at least one similarity measure;
    The identifying step 13 includes identifying at least one representative image 2, 5, 7 in each cluster,
    Said independent coding step 15 comprises independently coding said representative image 2, 5, 7 from each cluster,
    The predictive coding step 14 uses each representative image 2, 5, 7 from the cluster as a reference image, and each non-representative image 1, 3, 4, from each of the clusters. Predictive coding 6, 8).
    [3" claim-type="Currently amended] The method of claim 1,
    Each of the non-representative images 1, 3, 4, 6, 8,
    Partitioning the non-representative image into target blocks TB1 and TB2 of a plurality of pixels;
    For each of the target blocks TB1, TB2, the pixels of the corresponding target blocks TB1, TB2 are pixels of the same sized reference blocks RB2, RB5 in the reference image according to a defined search metric. Comparing the data, and producing an error metric for each comparison,
    Determining whether any value of the error metric values produced for each of the target blocks TB1, TB2 is less than a prescribed maximum threshold, and
    If less than a prescribed maximum threshold, identifying one of said equally sized reference blocks RB2, RB5 that constitutes the best pair for the target blocks TB1, TB2, and
    If not less than the defined maximum threshold, reference is made to the reference images 2, 5, 7 selected in the predictive coding step 14 by independently coding the target blocks TB1, TB2. How is it coded
    [4" claim-type="Currently amended] The method of claim 3, wherein
    The predictive coding step 14,
    The reference images 2, 5 constituting the best pair for the respective target blocks TB1, TB2 of each of the non-representative images 1, 3, 4, 6, 8, which are not independently coded. Identify reference blocks RB2, RB5 of 7), and which target blocks TB1, TB2 of each of the non-representative images 1, 3, 4, 6, 8 are independently coded; In addition, certain images of the set of images 1,..., 8 are reference images 2, 5, 7, and some images are non-representative images 1, 3, 4,. 6, 8) further comprising producing a coded data stream ({1 * , ..., 8 * }) identifying the identity.
    [5" claim-type="Currently amended] The method of claim 3, wherein
    The predictive coding step 14,
    For each of the target blocks TB1, TB2 that are not independently coded, to produce a residual value for each of the target blocks TB1, TB2 that are not independently coded, the corresponding target blocks TB1, TB2. ) And a step of calculating a difference between the reference blocks RB2 and RB5 constituting the best pair, and
    And a substep of coding the residuals produced in the calculation step of producing a coded data stream ({1 * , ..., 8 * }) representing the calculated residuals.
    [6" claim-type="Currently amended] The method of claim 2,
    Wherein the similarity measure comprises at least one similarity measure selected from the group consisting of color similarity measure, texture similarity measure, shape similarity measure, geometrical similarity measure, and content similarity measure.
    [7" claim-type="Currently amended] The method of claim 2,
    Partitioning the still images 1,..., 8 into a plurality of color comparison blocks,
    Calculating a standardized histogram for each of said color comparison blocks of each image, and
    Using the calculated and standardized histograms in performing the grouping and identification step.
    [8" claim-type="Currently amended] The method of claim 2,
    The method,
    Subdividing each still image 1,..., 8 into at least two compartments,
    The grouping step (13) is carried out by an image according to at least one similarity measure of the corresponding one of the compartments having a location corresponding to each image, to produce a set of clusters for each corresponding set of compartments. Grouping into at least two clusters based on similarity,
    The identifying step 13 identifies at least one representative image 2, 5, 7 for each cluster of compartments, wherein the other images 1, 3, 4, 6, 8 in each of the clusters are identified. ) Includes said identifying step, identified as a non-representative image,
    Said independent coding step 15 comprises coding a reference section in at least one representative image for each of said clusters,
    The predictive coding step 14 uses each reference in non-representative images 1, 3, 4, 6, 8 for each cluster using the reference partition in at least one representative image for the cluster as a reference image. Coding said partitions.
    [9" claim-type="Currently amended] The method of claim 1,
    Subdividing each image 1,..., 8 into at least two compartments,
    The method comprises the step (12) of compressing the sections rather than the set of still images (1, ..., 8).
    [10" claim-type="Currently amended] In the apparatus 12 for compressing a set of still images 1,..., 8,
    The device,
    Means (13) for identifying at least one representative image (2, 5, 7) in the set of still images (1, ..., 8), the other still images (1, 3, 4, 6, 8, said identification means 13, identified as a non-representative image,
    Means 15 for independently coding said at least one representative image 2, 5, 7, and
    Means for predictively coding each of said non-representative images (1, 3, 4, 6, 8) using said at least one representative image (2, 5, 7) as a reference image .
    [11" claim-type="Currently amended] In the transmitter 11,
    An apparatus 12 for compressing the set of still images of claim 10, and
    Means (16) for transmitting said set of compressed still images (1, ..., 8).
    [12" claim-type="Currently amended] For a coded data stream (1 *, ..., 8 *) representing a set of compressed still (1 * , ..., 8 * ) images,
    At least one representative still image 2, 5, 7 is independently coded 15, and the non-representative images 1, 3, 4, 6, 8 are reference images and at least one representative image 2, 5, 7) the data stream to be predictively coded with (14).
    [13" claim-type="Currently amended] Storage medium (17) in which a set of the compressed still images ({1 * , ..., 8 * } ') of claim 12 are stored.
    [14" claim-type="Currently amended] A method (32) for decompressing a set of compressed still images ({1 * , ..., 8 * }),
    The method,
    Identifying 34 at least one representative image 2 * , 5 * , 7 * in the set of compressed still images 1 * ,..., 8 * , all other compression in the set The identified step 34, wherein the captured still images 1 * , 3 * , 4 * , 6 * , 8 * are identified as non-representative images,
    Independently decoding 36 the at least one representative image 2 * , 5 * , 7 * , and
    Predictively decoding each of the compressed non-representative images 1 * , 3 * , 4 * , 6 * , 8 * using the at least one representative image 2, 5, 7 as a reference image ( 35).
    [15" claim-type="Currently amended] In the device 32 for decompressing a set of compressed still images 1 * ,..., 8 * ,
    The device 32,
    Means 34 for identifying at least one representative image 2 * , 5 * , 7 * in the set of compressed still images 1 * ,..., 8 * ; The still images 1 * , 3 * , 4 * , 6 * , 8 * are identified as the non-representative images, the identification means 34,
    Means 36 for independently decoding said at least one representative image 2 * , 5 * , 7 * , and
    Means 35 for predictively decoding each of the compressed non-representative images 1 * , 3 * , 4 * , 6 * , 8 * using the representative images 2, 5, 7 as reference images Containing device.
    [16" claim-type="Currently amended] In the display device 30,
    An apparatus 32 for decompressing the compressed still images 1 * ,..., 8 * of claim 15, and
    Apparatus comprising a display unit (33) for displaying the decompressed still images (1, ..., 8).
    类似技术:
    公开号 | 公开日 | 专利标题
    RU2687039C1|2019-05-06|Device for coding images with prediction, prediction image decoding apparatus, prediction image coding method, prediction image decoding method, program for coding prediction images and program for decoding prediction images
    CN104244007B|2020-09-01|Image coding method and device and decoding method and device
    US8582903B2|2013-11-12|Efficient macroblock header coding for video compression
    US20150281700A1|2015-10-01|Image encoding/decoding method and device
    Yue et al.2013|Cloud-based image coding for mobile devices—Toward thousands to one compression
    Qiu2003|Color image indexing using BTC
    US6611628B1|2003-08-26|Method of image feature coding and method of image search
    US6476805B1|2002-11-05|Techniques for spatial displacement estimation and multi-resolution operations on light fields
    US7085319B2|2006-08-01|Segment-based encoding system using segment hierarchies
    US6567561B2|2003-05-20|Video coding and video decoding apparatus
    JP3727341B2|2005-12-14|Scaleable pixel distribution image coding system for image compression and decompression
    US7949053B2|2011-05-24|Method and assembly for video encoding, the video encoding including texture analysis and texture synthesis, and corresponding computer program and corresponding computer-readable storage medium
    CA2294388C|2007-04-24|Method for generating sprites for object-based coding systems using masks and rounding average
    RU2654200C1|2018-05-17|Improved coding of the screen content with improved methods of palette and index card coding
    US6862371B2|2005-03-01|Method of compressing images of arbitrarily shaped objects
    US7082162B2|2006-07-25|Segment-based encoding system including segment-specific metadata
    US20140185949A1|2014-07-03|Efficient compact descriptors in visual search systems
    KR100452064B1|2004-10-08|Apparatus for processing digital video
    US7050503B2|2006-05-23|Segment-based encoding system using residue coding by basis function coefficients
    US6438168B2|2002-08-20|Bandwidth scaling of a compressed video stream
    US6888894B2|2005-05-03|Segmenting encoding system with image segmentation performed at a decoder and encoding scheme for generating encoded data relying on decoder segmentation
    US7006689B2|2006-02-28|Image search system and image search method thereof
    US6895051B2|2005-05-17|Video data encoder and decoder
    US5835034A|1998-11-10|System and method for lossless image compression
    CA2104287C|2003-02-18|Method and apparatus for encoding of data using both vector quantitization and runlength encoding and using adaptive runlength encoding
    同族专利:
    公开号 | 公开日
    CN1310913A|2001-08-29|
    US6625319B1|2003-09-23|
    CN1214616C|2005-08-10|
    WO2000060847A1|2000-10-12|
    EP1084568A1|2001-03-21|
    JP2002541738A|2002-12-03|
    引用文献:
    公开号 | 申请日 | 公开日 | 申请人 | 专利标题
    法律状态:
    1999-03-30|Priority to US09/281,353
    1999-03-30|Priority to US09/281,353
    2000-03-09|Application filed by 요트.게.아. 롤페즈, 코닌클리케 필립스 일렉트로닉스 엔.브이.
    2000-03-09|Priority to PCT/EP2000/002102
    2001-03-26|Publication of KR20010025113A
    优先权:
    申请号 | 申请日 | 专利标题
    US09/281,353|1999-03-30|
    US09/281,353|US6625319B1|1999-03-30|1999-03-30|Image compression using content-based image similarity|
    PCT/EP2000/002102|WO2000060847A1|1999-03-30|2000-03-09|Image compression|
    [返回顶部]