High dynamic range image sensor

专利摘要:
An image processing system has an array of sensors with a plurality of pixels. A separate circuit is associated with each pixel in the sensor array. The circuits are designed so that, on the plurality of circuits, there are a plurality of distinct sensitivities to light illumination of a scene to be captured, which are placed in pseudo-random spatial order with respect to each other. . The image processing system also includes an image reconstruction algorithm. A method is also described.
公开号:BE1023383B1
申请号:E2015/5538
申请日:2015-08-26
公开日:2017-03-01
发明作者:Penghe Geng；Hongcheng Wang；Alan Matthew Finn
申请人:Sensors Unlimited, Inc.；
IPC主号:

专利说明:

IMAGE SENSOR WITH HIGH DYNAMIC RANGE
Context of the invention
The present invention relates to an image acquisition and processing technique where the image is captured using a sensor that has different sensitivity levels assigned to different pixels.
Image reconstruction is inherent in any number of technical fields. For example, surveillance aircraft capture images at multiple wavelengths that need to be reconstructed to provide information. These images must be captured relatively quickly, and the accuracy, spatial resolution, and dynamic range must be as large as possible.
However, a natural scene usually has a very large dynamic range, namely, very bright and very dark areas, requiring for example 20 bits, and the standard imaging sensors can acquire less, for example only 8 to 12 bits. A conventional imaging sensor faces the problem of lack of scene detail or blur or severe distortion due to limited dynamic range.
Prior art methods for capturing high dynamic range ("HDR") images can use multiple sequential exposures to obtain multiple images at different exposures. Also, they can use multiple different sensor networks with different sensitivities. In addition, they can manufacture a single chip with multiple pixels of different sizes to simultaneously capture multiple images with different exposures. In yet another approach, they have attempted to integrate a luminous flux until a pixel reaches saturation where the integration time represents the actual illuminance. An additional technique is to use pixels or logarithmic response circuits to nonlinearly extend the dynamic range of a scene. Finally, masks or physical filters were used, with pixel attenuation levels, deployed in regular diagrams.
With these techniques, the final image is obtained for display by compressing the large dynamic range constructed from the sampling data.
All approaches of the prior art have disadvantages. The approach of taking multiple sequential exposures presents artefacts due to motion in the scene, as each image is of a slightly different time or a slightly different duration. In addition, a superposition of complicated images is necessary to generate the final image. There are also often artifacts in reconstructed images due to occlusion or improper overlay.
The many different sensor arrays present the problem that light is split and reflected back to different sensor arrays. The amount of light reaching a sensor is therefore lower, and this is not appropriate for low light level imaging. The additional equipment also increases the cost, size, weight and electrical energy required. The use of multiple size pixels reduces the resolution relative to a single size pixel and thus is also not ideal.
In the time-integration system of pixels, a main disadvantage is that a partially open transfer transistor may introduce an additional source of dark current resulting in a higher dark current shot noise. Also, the excitation circuits are more complex because of multiple signal readings. Additional electronic components are needed at a proportionately higher cost.
The pixel or logarithmic response circuitry has a non-linear response that is not preferred in most applications because it makes it difficult to accurately correct colors over the full range of sensor performance.
The mask system has generally been provided with a regular grid mask, resulting in low resolution or poor quality due to imaging interpolation.
Thus, an improved high dynamic range imaging method is desirable. Summary of the invention
An image processing system has a sensor array with a plurality of pixels. A separate circuit is associated with each pixel in the sensor array. The plurality of circuits is designed such that, over the plurality of circuits, there are a plurality of distinct light-illuminating sensitivities of a scene to be captured, which are spatially-pseudo-random relative to each other. to others. The image processing system also includes an image reconstruction algorithm. A method is also described.
To best understand these features, and others, the following drawings and brief are provided.
Brief description of the drawings
Figure 1 shows a pseudo-random sensor.
Figure 2 is a diagram of a system.
Fig. 3A shows a first circuit embodiment.
Fig. 3B shows a second circuit embodiment.
Figure 3C illustrates another way of providing distinct sensitivity levels and circuits.
Figure 4 graphically represents a step in a dictionary learning algorithm.
Figure 5 shows typical images.
detailed description
The present invention relates to capturing a pseudo-random sensor image. The following description is merely illustrative in nature and is not intended to limit the invention, its application, or its uses.
In Figure 1, a sensor 40 is shown schematically. A plurality of pixels in an array of w-by-w pixels is provided. Each pixel in the sensor communicates with a circuit, and the circuits have different sensitivities to illuminance. Thus, some pixels 22 are associated with circuits 225 and are sensitive to low light illumination (have a high sensitivity) while other pixels 23 are associated with circuits 235 and are sensitive to high light illumination (have a low sensitivity). Although only one of each circuit 225 and 235 is illustrated, and schematically illustrated, each pixel is associated with a circuit. Thus, there will be a plurality of the two circuits 225 and 235.
As can be appreciated from FIG. 1, the different sensitivities are irregularly spaced or are not placed in a geometric order, in correspondence with a spatial irregularity on the sensor 40. This property, not placed in spatial order, is called a pseudo diagram. -random. These sensitivities are provided by the circuit that communicates with each pixel and communicates the pixel information to a downstream processor or computer (best described below with reference to Fig. 2).
As shown in FIG. 2, a camera 58 comprising the sensor 40 can communicate with a processor or computer 60. The scene 32, intended to be captured, is represented reconstructed and displayed, at reference 132, on the processor or the computer 60. The use of the pseudorandomized sensor provides advantages, as will be described below. Particularly, when used in conjunction with modern reconstruction algorithms, more accurate images are provided.
Although the processor or the computer 60 is shown separately from the camera 58, which itself comprises the sensor 40, the processor or the computer 60 can be housed inside the camera. , or otherwise integrated with, or incorporated with, the camera 58 or even even inside the sensor 40. The communication between the camera 58 and the computer 60 may be a cable channel comprising conductors on an integrated circuit or alternatively may be any other optical, wireless, radio, or other type of channel capable of transmitting images and videos between two points, including links using the World Wide Web (" World Wide Web "or www) or the Internet.
The present invention describes the acquisition and reconstruction of high quality images from a single exposure using a single sensor 40. Standard sensors will not generate high quality images as high as the pseudo-random sensor 40 and the reconstruction techniques of the present invention. Multidimensional sampling, using the techniques of the present invention, can be used to obtain a high resolution image, a low cost, a high speed, and a large dynamic range.
Pseudo-random imaging reconstruction depends on a mathematical property called parsimony. Parsimony is a property by virtue of which some data, for example, an image, can be represented only by some non-harmful numbers (also called coefficients) that multiply an appropriate set of basic functions. We know that natural imagery is parsimonious because these images can be compressed (for example, using relatively few coefficients and Fourier-based functions or wavelet) and reconstructed with precision from these few coefficients.
The present invention creates a pseudo-randomized sensor of several distinct levels of different sensitivity where each pixel of an imaging chip communicates with one of the circuits 225, 235. An image of a natural scene is captured by the intermediate sensor. The pixels corresponding to a sensitivity level are called pseudo-random image. For each exposure, then, there are several pseudo-random images each corresponding to the pixels of each sensitivity level. From the acquired pseudo-random images, computational methods are used to reconstruct the desired high dynamic range (HDR) image.
A sensor with N distinct levels of different sensitivity is created. There are, therefore, N pseudo-random images for any scene corresponding to N levels. The darkest pseudo-random image includes the pixels sensitive to the lowest illuminance (those with the highest sensitivity). Similarly, the next darkest pseudorandom image comprises the pixels of the next lowest sensitivity level, and so on. For each exposure, then, there are N pseudorandom images each corresponding to the pixels of a sensitivity level. For each pseudo-random image (corresponding to each sensitivity level), there is a limited number of observed pixels (l / N), from which the entire image must be reconstructed.
In the illustrative embodiment, each pixel of a sensor comprising n-by-m pixels communicates with a circuit 25, 235 of a sensitivity level.
The method of producing a sensor with a pseudo-random spatial distribution of sensitivities may vary. In one example, different levels of sensitivity are assigned independently, at each pixel location, and uniformly throughout the sensor. In more general cases, levels can be correlated and evenly distributed on average. The amplitude of the spatial variation of levels in a subset of adjacent locations is the key parameter of a distribution that can control the statistical properties of the sensor. This illustrative distribution depends on the number of locations in the subset and a correlation function. Specifically, in the first case of independently distributed levels, the square amplitude of spatial variation can be directly proportional to the number of levels in the subset. In the second case of correlated distributions, this dependency can be modified. One of the most common examples is a power law dependency where the standard deviation is proportional to N 7, where N is the number of levels in a subset and y is a chosen parameter. For image acquisition and image processing applications, the correlation, and thus the parameter y, can be optimized to reduce the local nonuniformity of the sensor and thereby increase the quality of image reconstruction.
In one application, FIG. 3A, the circuit 21 provides a gain component 122 which can be fixed or can be varied in a programmable manner. Figure 3A shows a silicon read-out integrated circuit (ROIC) which has the ability to continuously change the analog gain to a pixel. By modifying internal polarizations, the gate-modulated (GMOD) pixel architecture ROIC can change the sensitivity of each pixel of the sensor (focal plane array), resulting in an effective capacitance change . Each pixel contains a gate modulated input circuit to convert current into voltage with continuously adjustable gain. The photodiode current passes through MO with a proportional amount of current reflected in M1 while the ratio of the currents by M1 and MO is controlled via externally adjusted gain ("GAIN") and polarization ("BIAS") voltages. ). Again, one of ordinary skill in the art will recognize the manner in which these variations are obtained using the present invention.
FIG. 3B is a prior art embodiment of a side-pass capacitor HDR circuit of FIG. 2.16 by Yang Liu, "The Design of a High Dynamic Range CMOS Image Sensor in the Technology," Master of Science Thesis, Delft University of Technology, 2012 which is hereby incorporated by reference. Figure 3B shows a pair of T2 and T3 transistor switches (17).
A second embodiment, FIG. 3C, shows a circuit in which a transistor switch 17 is eliminated from FIG. 3B and the overflow capacitor CS (16) is modified in the form CS '(16'). One of ordinary skill in the art will recognize that by changing the capacitance of the overflow capacitor 16 ', the sensitivities of each pixel can be varied. For example, by configuring the overflow capacitor CS '(16') with different capacitance levels, different levels of sensitivity for each pixel can be obtained. For high sensitivity pixels, the overflow capacitor CS 'can be eliminated completely. Thus, single circuits could be pseudo-randomly provided on all of the pixels of the sensor 40. In addition, with the modifications of the present invention, the designs of the image sensor pixel and the read circuit are simplified.
In addition, the effective charge storage of a pixel can be changed by adjusting the Floating Diffusion (FD) capacitance (18) so that the high sensitivity pixels have a lower capacitance and the pixels to low sensitivity have a higher capacitance. By changing the capacitance corresponding to each pixel, its sensitivity level can be changed. For example, capacitance can be changed by increasing the area, increasing the thickness of the deposit layer, and so on. A possible example is described in US Patent 5,621,230, the relevant disclosure of which is hereby incorporated by reference.
Although two different levels of sensitivity are shown in Figure 1, it should be understood that three or more levels of sensitivity may also be included.
In summary, the various circuits, as shown in Figs. 3A to 3C, result in an integrated readout circuit (ROIC), wherein the circuits provided for each of the pixels have a plurality of distinct levels of sensitivity to illumination. luminous.
A reconstruction algorithm is then used. One embodiment may use a well-known low rank matrix completion algorithm. Another is the application of a dictionary learning algorithm for image painting. The preferred embodiment uses algorithms based on li / total variation minimization ("Total Variation" or TV). The basic concepts of li and TV minimization are well known in the art and are explained further below. The different scene components can be reconstructed independently (this is called independent reconstruction) or, preferably, by joint optimization (this is called joint reconstruction). The independent reconstruction approach reconstructs each scene component independently, using only pixel responses corresponding to a sensitivity level. The joint reconstruction approach reconstructs all the components of the scene at the same time, implicitly or explicitly assuming that the structure of the scene components is linked and using the pixel responses corresponding to a plurality of sensitivity levels. The independent reconstruction algorithms are well known in the art. The new joint reconstruction algorithm below shows better reconstruction accuracy than independent reconstruction.
The present invention comprises the following three steps: 1) Pseudo-random image acquisition: A conventional image sensor with a pseudo-random circuit takes a single exposure acquiring a plurality of pseudo-random images. 2) Image reconstruction, for example, with dictionary learning or li / TV based approaches, as explained below: scene components are reconstructed independently or together. From the acquired pseudorandom images, there are two methods that can be used to reconstruct the desired high dynamic range (HDR) image. One is a process based on local parcels, namely, image reconstruction based on dictionary learning. The other is a global image-based method, namely a li / TV based image reconstruction. These are described in detail below. 3) High dynamic range compression: Finally, the reconstructed images are combined to produce a single image with a large dynamic range. The large dynamic range can optionally be compressed for display.
Theoretical mathematical developments over the last decade in parsimonious sampling and parsimonious optimization (also known as "compressive sensing" and "inverse regularization problems") have presented new ways of recovering missing information from data. sampled appropriately. Proper sampling requires some sample pseudo-randomness in order to function properly.
The regularity of the sampling (spatial distribution of sensitivity levels) limits the successful use of these new mathematical developments. Spatially pseudo-random sampling of a scene at a particular sensitivity level allows us to accurately reconstruct the entire image as if the entire image were acquired at that level.
Natural scenes contain many spatially regular structures, for example, windows on an office building, fence posts, and so on. If a scene is sampled in a regular mode, the regularity of the samples may cause problems in the reconstruction of the image. The moiré, which can be seen on an actor's clothes on television, is a well-known example. In this case, regular spatial sampling of a regular spatial pattern on clothing may result in reconstruction artifacts due to the well known aliasing effect. Another related known example is the temporally regular sampling of a rotating wheel which may give it the appearance of alternately moving forwards and backwards as it accelerates or slows down. As an extreme example, regular sampling of a scene with a fence could only include samples of pickets (whose reconstruction would create the image of a solid wall) or only samples of pickets (of which the reconstruction would create an image without any fence). The reason that pseudo-random sampling is effective and that obtaining information, with respect to any regular structure, sufficient to allow accurate reconstruction is much more likely. The idea of dictionary learning is to learn a compact dictionary from the sampled pseudo-random image to reconstruct the high resolution image. A dictionary (indicated by Φ, also called a sampling matrix or detection matrix) for an image, x, allows an accurate reconstruction provided that the following two conditions are satisfied: (1) Parsimony: the mathematical representation of the image , Φχ, is parsimonious, given a dictionary Φ too complete and redundant (redundancy here means that the number of dictionary atoms is much larger than the dimension of image parcels of x, which implies that Φχ contains many zeros). As mentioned above, parsimony is a property by which an image can be represented by only a few nonzero numbers (also called coefficients) that multiply an appropriate set of basic functions (each basic function is a vector called an atom, the collection of atoms forms a dictionary in the form of columns of the dictionary). (2) Inconsistency: the detection matrix / measurement matrix ΦT has a complete star ("spark"). The star of a dictionary (matrix) is the lowest number of columns that are linearly dependent. The complete star means that no square submatrix of the matrix is singular. If columns are linearly dependent then they will not add any new information to the sampling process. The star is useful in the theory of compressive detection, where the necessities on the star of the measurement matrix 0 ^ are used to guarantee the stability and constancy of the mathematical techniques. A related measure of inconsistency between dictionary atoms is the property of well-known Restriction Isometry Property (RIP).
The pseudo-random character of the spatial distribution of sensitivities is important to guarantee the inconsistency of the detection matrix Φ ^ A regular grid spatial distribution of sensitivities will present linear dependences between dictionary atoms and therefore ΦT has a non-star complete and has a worse RIP than for a pseudo-random spatial distribution of sensitivities. Dictionary-based reconstruction results using a spatial spatially regular spatial distribution of sensitivities are much worse than when a pseudo-random spatial distribution of sensitivities is used. Similarly, for lj / TV based approaches, the pseudo-random sampling matrix (indicated by P, below) is the detection matrix, which has a good RIP.
The dictionary-based image reconstruction uses image parcels, see Figure 4. First, a pseudo-random image is divided into a set of overlapping parcels, x ,, each of a size. axa (for example, a = 8). (Hereinafter, plot x, is considered to be a column vector as with the well-known Matlab command vect (x;).) An illustrative plot 100 is shown in Figure 4. The parsimony constraint is exploited in that each parcel of the image is represented as a parsimonious combination of a set of dictionary atoms. Images can be reconstructed from an individual dictionary learned from each pseudo-random image, or from a single dictionary learned from all pseudo-random images. Dictionary learning is described as follows.
so that
where Xi are image parcels, Φ is the dictionary, ", are parsimonious coefficients, and τ0 is a weak constant. It should be noted that the constraint 10 of parsimony
has in fact been expressed by the equivalent constraint
as is well known from the compressive detection documentation. The intuitive interpretation of this optimization problem is that a dictionary Φ and coefficients a are computed so that the sum of the differences between the parcels of images x, and their approximation from a dictionary, Φα, is small (each individual plot difference is the term
which measures the difference that the plot differs from its parsimonious dictionary representation). The notation
is a measure of difference, namely, a Euclidean distance (squared) between two vectors. Summation
add up all individual plot differences.
At the same time as the parcel differences are minimized, the guarantee of parsimony of representation (this is the term
), which requires that the parsimony of a is smaller than a low number τ0 that is specified, is also desired. The notation
is the parsimony measure (also called 10), an count of the number of nonzero elements of a vector, which has been replaced by its equivalent (in this case)
(also called li).
Thus, the solution of this optimization problem generates a dictionary that can represent all the image parcels where each parcel representation requires only a few dictionary elements. The mathematical theory ensures that if this dictionary is computed, the entire image can be reconstructed even if only l / N of the actual pixel values are available. The learning dictionary continues as follows.
The dictionary, Φ, can initially be set to any value or to the singular value decomposition (SVD) that is well known in all parcels. Learning a dictionary has two main steps: parsimonious encoding step: for each parcel xh a parsimonious representation, ai, is computed using any tracking algorithm (for example, the well-known basic tracking algorithm) so that each x is a combination of a parsimonious set of dictionary atoms; dictionary update step: each atom of the dictionary Φ is updated as the first eigenvector of the error matrix from the parsimony adjustment for the parcel group using this atom.
Both steps are repeated until convergence. This procedure is well known in the documentation.
The independent reconstruction technique described next is the 1 / TV based image reconstruction. This technique imposes parsimony on an entire image (rather than parcels) insofar as any natural image can be represented as a parsimonious number of coefficients according to a certain basis (for example, Fourier, or wavelets), that is, a constraint lb or it can be represented as a gradient field to parsimonious piece constant, i.e., a TV constraint. Images are reconstructed from pseudorandom images acquired independently or jointly.
Independent reconstruction has the formulations l! and TV following: wording 1:
so that
where x is the image to reconstruct, F is the inverse-based transformation (for example, Fourier, wavelets), P is the subsampling operator corresponding to the pseudo-random subsampling in pseudo-random image b, and δ is a low number that we choose. TV formulation:
so that
or
is the total variation, and P is the subsampling operator corresponding to the pseudo-random subsampling in pseudo-random image b, and δ is a low number that one chooses.
This is an independent reconstruction approach that does not use known relationships in pseudo-random images at different levels of sensitivity. In an independent reconstruction, each pseudo-random image is reconstructed separately and then later combined into a single HDR image. It is known, however, that successive pseudo-random images, corresponding in reality to the same scene imaged at different moments of exposure, must be strongly related to each other. An innovative joint reconstruction approach that simultaneously uses information from all pseudorandom images can use more relationships and thus achieve better reconstruction accuracy than independent reconstruction.
To use the relationships between pseudorandom images, the pseudorandom images are modeled with some physical imaging constraint, for example, the well-known camera response function. For example, for the High Dynamic Range Imaging (HDR) application, images are actually acquired with different exposure times for different pixels, so that the response function of the camera ( "Camera Response Function" (CRF) is used to model the illuminance value with respect to exposure time. The imaging model can be represented as xt = / (log (<5f.)), Where <5 lt is the duration of exposure, and ƒ is the estimated camera response function from the pseudorandom images acquired or calculated a priori.
Using the same notation as this one, the joint reconstruction is formulated as follows: formulation lj:
so that
where n is the number of images to reconstruct, TV formulation:
so that
For the TV formulation, the well-known Split-Bregman iteration approach is used to efficiently reconstruct the images by the following three steps: (1) Application of the Bregman formulation by introducing auxiliary variables. (2) Decoupling portions lj and 12 of the new cost function. (3) Solving the minimization of each cost function in turn to convergence, solving a classical Sylvester equation and a shrinkage problem.
The present invention advantageously uses the pseudo-random distribution of the spatial distribution of sensitivities. In particular, the image reconstruction described is based on the resolution of an optimization problem (typically 12/1 / mixed standard optimization). A key necessity of this type of optimization problem is that spatial (spatial-temporal) sampling is pseudo-random - specifically, it has the full star or a good RIP. The pseudo-randomness can come from any one of a number of underlying pseudo-random number distributions. The pseudo-random spatial distribution of sensitivities can also be optimally designed to have a better star or RIP property.
Figure 5 shows three images of distinct levels of sensitivity 110, 112 and 114 that can be captured from a scene. As shown, an associated sensor 140 has three distinct groups of pixels, 139 being the least sensitive. The pixels 141 are an intermediate level. The pixels 142 are the most sensitive.
Thus, pixels 139 and circuits 143 will capture darker areas such as the interior of the building, image 114. Pixels 141 and circuits 144 will capture the more intermediate levels, as shown on 112. Pixels 142 and circuits 145 are best able to capture bright areas, such as the exterior of the building, image 110. When these multiple images are reconstructed, as described, a very high resolution image 116 is obtained.
In sum, the use of pseudo-random circuits provides a higher resolution in the final recombined image.
It is not necessary that the term "pseudo-random" as used herein be actually pseudo-randomly produced. Specifically, the pseudo-random distribution may be truly random or may be approximately random, as produced by any number of techniques, for example, optimized methods of spatial correlation. It is essential that the spatial distribution of sensitivities is not put in regular order.
The present invention for HDR imaging first estimates the CRF from the acquired images. The CRF is then used as part of the mixed standard optimization. Reconstructing a single high dynamic range (HDR) image from multiple images at different levels of exposure using a CRF is known in the art.
Although embodiments of the present invention have been described, those skilled in the art will recognize that certain modifications will be within the scope of the present invention. For this reason, the following claims must be studied to determine the true scope and true content of the present invention.

权利要求:
Claims (15)
[1]
A high dynamic range imaging system, comprising: a sensor array with a plurality of pixels; and a plurality of circuits, with a circuit associated with each pixel in said array of sensors, said plurality of circuits being arranged so that, on said plurality of circuits, there is a plurality of distinct sensitivities to a luminous illumination of a scene intended to be captured, where said sensitivities are placed in pseudo-random spatial order with respect to each other; and an image reconstruction algorithm.
[2]
The system of claim 1, wherein said sensor array communicates with a computer provided with said reconstruction algorithm to reconstruct one or more distinct images from one or more of said distinct qualities of the scene.
[3]
The system of claim 2, wherein said computer is an embedded processor.
[4]
The system of claim 2, wherein said discrete images are combined into a single combined image.
[5]
The system of claim 1, wherein said reconstruction algorithm uses one or more of dictionary training, 1 i / total variation based optimization, and matrix completion.
[6]
The system of claim 1, wherein said plurality of circuits comprises certain circuits having a photodiode with a first overshoot capacitance and other circuits having a photodiode with a second lower overshoot capacitance to provide said distinct sensitivities to the illuminance.
[7]
The system of claim 1, wherein a gain in each of said plurality of circuits is varied to provide said distinct sensitivities to illuminance.
[8]
The system of claim 1, wherein the pseudo-random ordering is a truly random, approximately randomly produced computer program, or approximately randomly optimized by spatial correlation.
[9]
A high dynamic range imaging method, comprising the step of: including a plurality of circuits associated with each pixel in a sensor array, said plurality of circuits being such that, on said plurality of circuits, there is a plurality of distinct sensitivities to a luminous illumination of a scene to be captured, where said sensitivities are placed in pseudo-random spatial order with respect to one another to form a plurality of distinct images, and reconstruction of the plurality of separate images to form an image.
[10]
The method of claim 9, wherein said sensor array communicates with a computer that reconstructs one or more distinct images from one or more of said distinct qualities of said scene.
[11]
The method of claim 10, wherein said computer is an embedded processor.
[12]
The method of claim 9, wherein said reconstructing uses one or more of a dictionary training, a 1 i / total variation based optimization, and a matrix completion.
[13]
The method of claim 9, wherein the pseudo-random ordering used is a truly random, approximately randomly produced by computer program, or approximately randomly optimized by spatial correlation.
[14]
The method of claim 9, wherein a gain in said plurality of circuits is varied to provide said distinct sensitivities to illuminance.
[15]
The method of claim 9, wherein said plurality of circuits comprise certain circuits having a photodiode with a first overshoot capacitance and other circuits having a photodiode with a second lower overshoot capacitance to provide said distinct sensitivities to the illuminance.

类似技术:

---

公开号 | 公开日 | 专利标题

---

BE1023383B1|2017-03-01|High dynamic range image sensor

---

Serrano et al.2016|Convolutional sparse coding for high dynamic range imaging

---

Costantini et al.2004|Virtual sensor design

---

BE1021668B1|2015-12-23|A PIXEL WITH A LARGE DYNAMIC RANGE AND A METHOD FOR OPERATING IT

---

EP2846533B1|2016-02-17|Device for compressive image acquisition

---

FR2959013A1|2011-10-21|DEVICE FOR DETECTING ELECTROMAGNETIC RADIATION WITH REDUCED SENSITIVITY TO SPACIAL NOISE

---

EP2833622B1|2015-11-04|Diagnosis of the faulty state of a bolometric detection matrix

---

Elgendy et al.2017|Optimal threshold design for quanta image sensor

---

FR2996034A1|2014-03-28|Method for generating high dynamic range image representing scene in e.g. digital still camera, involves generating composite images by superposition of obtained images, and generating high dynamic range image using composite images

---

Guicquero et al.2016|An algorithm architecture co-design for cmos compressive high dynamic range imaging

---

FR2976121A1|2012-12-07|MATRIX IMAGING DEVICE COMPRISING AT LEAST ONE SET OF MULTIPLE TIME INTEGRATION PHOTOSITES.

---

US8958649B2|2015-02-17|Video generation with temporally-offset sampling

---

Ginhac et al.2010|A high speed programmable focal-plane SIMD vision chip

---

Fereyre et al.2011|L² CMOS image sensor for low light vision

---

WO2018020115A1|2018-02-01|Microwave radiometry imaging system and associated imaging method

---

FR2945667A1|2010-11-19|INTEGRATED IMAGE SENSOR WITH VERY HIGH SENSITIVITY.

---

FR3062009B1|2019-08-16|ADAPTIVE GENERATION OF A DYNAMICALLY ENHANCED SCENE IMAGE OF A SCENE FROM A PLURALITY OF IMAGES OBTAINED BY NON-DESTRUCTIVE READING OF AN IMAGE SENSOR

---

Ghasemi et al.2017|CMOS approach to compressed-domain image acquisition

---

WO2014195622A1|2014-12-11|Cmos sensor with standard photosites

---

FR2968876A1|2012-06-15|System for acquisition of images of scene, has separation unit separating flow of beam from focusing optics into predetermined fractions, and merging unit merging images output from sensors to generate final image

---

WO2021104905A1|2021-06-03|Image capture using radiation-sensitive elements having a memory effect

---

Feng2012|Bits from photons: Oversampled binary image acquisition

---

Guicquéro2015|Exploring information retrieval using image sparse representations

---

Yu et al.2015|IMRAN: a noise estimation method for relative radiometric calibration data

---

Saffih et al.2004|Human perception of fixed pattern noise in pyramidal CMOS image sensor

同族专利:

---

公开号 | 公开日

---

BE1023383A9|2017-03-24|

---

TWI668998B|2019-08-11|

---

US9467628B2|2016-10-11|

---

BE1023383A1|2017-03-01|

---

US20160065824A1|2016-03-03|

---

TW201608891A|2016-03-01|

引用文献:

---

公开号 | 申请日 | 公开日 | 申请人 | 专利标题

---

US20120086781A1|1998-09-28|2012-04-12|Microsoft International Holdings B.V.|3D Vision On A Chip|

---

US20030197793A1|2001-01-09|2003-10-23|Tomoo Mitsunaga|Imaging device|

---

EP1255410A2|2001-05-02|2002-11-06|Agilent Technologies, Inc. |System and method for capturing color images that extends the dynamic range of an image sensor|

---

WO2013089036A1|2011-12-16|2013-06-20|ソニー株式会社|Image pickup device|

---

US7180588B2|1999-04-09|2007-02-20|Plain Sight Systems, Inc.|Devices and method for spectral measurements|

---

US6864916B1|1999-06-04|2005-03-08|The Trustees Of Columbia University In The City Of New York|Apparatus and method for high dynamic range imaging using spatially varying exposures|

---

US6922210B2|1999-07-30|2005-07-26|Pixim, Inc.|Memory updating for digital pixel sensors|

---

US7542133B2|2003-12-22|2009-06-02|Searete, Llc|Photo-detector filter|

---

US20060103749A1|2004-11-12|2006-05-18|Xinping He|Image sensor and pixel that has switchable capacitance at the floating node|

---

JP2006148350A|2004-11-17|2006-06-08|Olympus Corp|Imaging apparatus|

---

WO2006116134A2|2005-04-21|2006-11-02|William Marsh Rice University|Method and apparatus for compressive imaging device|

---

JP2009516476A|2005-11-14|2009-04-16|サーノフコーポレーション|CMOS image sensor with hybrid pixel array|

---

US20080198235A1|2007-02-16|2008-08-21|Shou-Lung Chen|High dynamic range image recorder|

---

US8445828B2|2010-07-01|2013-05-21|Silicon Optronics, Inc.|High dynamic range image sensor with in pixel memory|US11178349B2|2015-01-29|2021-11-16|William Marsh Rice University|Lensless imaging system using an image sensor with one or more attenuating layers|

---

US10048413B2|2016-06-07|2018-08-14|Goodrich Corporation|Imaging systems and methods|

---

US11184564B2|2017-04-28|2021-11-23|The Governing Council Of The University Of Toronto|Method and system for pixel-wise imaging|

---

CA3061622A1|2017-04-28|2018-11-01|The Governing Council Of The University Of Toronto|Method and system for pixel-wise imaging|

---

US10657446B2|2017-06-02|2020-05-19|Mitsubishi Electric Research Laboratories, Inc.|Sparsity enforcing neural network|

---

KR102314703B1|2017-12-26|2021-10-18|에스케이하이닉스 주식회사|Joint dictionary generation method for image processing, interlace based high dynamic range imaging apparatus using the joint dictionary and image processing method of the same|

---

US10546369B2|2018-01-09|2020-01-28|Omnivision Technologies, Inc.|Exposure level control for high-dynamic-range imaging, system and method|

---

US10488371B1|2018-05-04|2019-11-26|United Technologies Corporation|Nondestructive inspection using thermoacoustic imagery and method therefor|

---

US10685433B2|2018-05-04|2020-06-16|Raytheon Technologies Corporation|Nondestructive coating imperfection detection system and method therefor|

---

US11079285B2|2018-05-04|2021-08-03|Raytheon Technologies Corporation|Automated analysis of thermally-sensitive coating and method therefor|

---

US10902664B2|2018-05-04|2021-01-26|Raytheon Technologies Corporation|System and method for detecting damage using two-dimensional imagery and three-dimensional model|

---

US10914191B2|2018-05-04|2021-02-09|Raytheon Technologies Corporation|System and method for in situ airfoil inspection|

---

US11268881B2|2018-05-04|2022-03-08|Raytheon Technologies Corporation|System and method for fan blade rotor disk and gear inspection|

---

US10958843B2|2018-05-04|2021-03-23|Raytheon Technologies Corporation|Multi-camera system for simultaneous registration and zoomed imagery|

---

US10943320B2|2018-05-04|2021-03-09|Raytheon Technologies Corporation|System and method for robotic inspection|

---

US10928362B2|2018-05-04|2021-02-23|Raytheon Technologies Corporation|Nondestructive inspection using dual pulse-echo ultrasonics and method therefor|

---

US10473593B1|2018-05-04|2019-11-12|United Technologies Corporation|System and method for damage detection by cast shadows|

法律状态:

优先权:

---

申请号 | 申请日 | 专利标题

---

US14/468,842|US9467628B2|2014-08-26|2014-08-26|High dynamic range image sensor|

---

US14/468,842|2014-08-26|

---