WO2014083857A1 - 画像処理装置、及び、画像処理方法 - Google Patents
画像処理装置、及び、画像処理方法 Download PDFInfo
- Publication number
- WO2014083857A1 WO2014083857A1 PCT/JP2013/007001 JP2013007001W WO2014083857A1 WO 2014083857 A1 WO2014083857 A1 WO 2014083857A1 JP 2013007001 W JP2013007001 W JP 2013007001W WO 2014083857 A1 WO2014083857 A1 WO 2014083857A1
- Authority
- WO
- WIPO (PCT)
- Prior art keywords
- constraint
- image
- regularization term
- weight
- reconstruction
- Prior art date
Links
- 238000003672 processing method Methods 0.000 title claims description 17
- 238000000034 method Methods 0.000 claims abstract description 61
- 230000008569 process Effects 0.000 claims abstract description 27
- 238000004364 calculation method Methods 0.000 claims description 196
- 238000013138 pruning Methods 0.000 claims description 21
- 238000006731 degradation reaction Methods 0.000 abstract description 3
- 230000015556 catabolic process Effects 0.000 abstract 1
- 239000011159 matrix material Substances 0.000 description 38
- 238000010586 diagram Methods 0.000 description 16
- 230000006870 function Effects 0.000 description 15
- 238000005516 engineering process Methods 0.000 description 6
- 230000008859 change Effects 0.000 description 5
- 230000000694 effects Effects 0.000 description 4
- 230000009467 reduction Effects 0.000 description 4
- 238000000611 regression analysis Methods 0.000 description 4
- 230000007423 decrease Effects 0.000 description 2
- 238000012804 iterative process Methods 0.000 description 2
- 230000002146 bilateral effect Effects 0.000 description 1
- 230000000593 degrading effect Effects 0.000 description 1
- 230000006866 deterioration Effects 0.000 description 1
- 230000004069 differentiation Effects 0.000 description 1
- 238000003384 imaging method Methods 0.000 description 1
- 239000004973 liquid crystal related substance Substances 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000000717 retained effect Effects 0.000 description 1
- 238000005070 sampling Methods 0.000 description 1
- 239000007787 solid Substances 0.000 description 1
- 238000010129 solution processing Methods 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T3/00—Geometric image transformations in the plane of the image
- G06T3/40—Scaling of whole images or parts thereof, e.g. expanding or contracting
- G06T3/4053—Scaling of whole images or parts thereof, e.g. expanding or contracting based on super-resolution, i.e. the output image resolution being higher than the sensor resolution
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04N—PICTORIAL COMMUNICATION, e.g. TELEVISION
- H04N1/00—Scanning, transmission or reproduction of documents or the like, e.g. facsimile transmission; Details thereof
- H04N1/40—Picture signal circuits
- H04N1/407—Control or modification of tonal gradation or of extreme levels, e.g. background level
Definitions
- the present invention relates to image processing, and more particularly, to an image processing apparatus and an image processing method for increasing the resolution of an image.
- the image (image) photographed by the photographing device has moved from the image on the film medium to digital data.
- Digital images are easier to process than film media images. For this reason, digital data images are processed using various image processing apparatuses.
- the super-resolution processing is processing for generating an image with a high resolution (high resolution image) based on an image with a low resolution (input image).
- the first super-resolution technique is a reconstruction-based super-resolution technique described in Patent Document 1.
- the resolution of an input image is reduced through a degradation process (for example, blur or low resolution).
- the reconstruction type super-resolution technique reconstructs the degradation process for the input image and generates a high-resolution image.
- the reconstruction type super-resolution technique can calculate a plurality of high-resolution image candidates (or solutions). Therefore, the reconstruction-type super-resolution technique uses constraint conditions in order to narrow down candidates (solutions) of the reconstructed high-resolution image.
- Reconfigurable super-resolution technology can set multiple constraints.
- the first constraint condition is a constraint condition based on reconstruction (hereinafter referred to as “reconstruction constraint”). More specifically, the reconstruction constraint is based on the relationship between the input image and the reconstructed high-resolution image (for example, the difference between the image obtained by degrading the high-resolution image and the input image) and the amount of reconstruction processing. This is a restriction to be determined based on this.
- the second constraint condition is a constraint based on the appropriateness of the high resolution image itself. More specifically, this constraint is a constraint based on the regularity of the matrix of the high-resolution image (hereinafter, this constraint is referred to as “regularization term constraint”). For example, the reconstruction type super-resolution technique adds a regularization term based on Bayesian theory (prior probability: prior probability). In this case, the regularization term constraint is a constraint that is determined based on the added regularization term.
- the reconstruction super-resolution technique narrows down the candidates (solutions) of the high-resolution image based on the reconstruction constraint and the regularization term constraint.
- the reconstruction-type super-resolution technique may be determined so that each constraint is calculated as a cost and the total cost is minimized as an actual process.
- the second super-resolution technique is a learning-based super-resolution technique.
- the learning type super-resolution technique creates a dictionary in advance based on the relationship between the learning high-resolution image and the low-resolution image.
- the learning super-resolution technique is a technique for generating a high-resolution image from an input image using the dictionary.
- the dictionary may be referred to as prior knowledge. Therefore, it can be said that the learning-type super-resolution technique uses prior knowledge as a constraint condition of the solution.
- the reconstruction super-resolution technology uses reconstruction constraints calculated using the input image.
- the learning type super-resolution technique uses a dictionary created based on the relationship between the learning high-resolution image and the low-resolution image.
- the image processing device replaces the reconstruction constraint of the reconstruction-type super-resolution technology with processing based on a dictionary of the learning-type super-resolution technology, and even if the regularization term constraint of the reconstruction-type super-resolution technology is used, An image can be generated. Therefore, in the following description, unless otherwise specified, the reconstruction-type super-resolution technique is described as a super-resolution technique including a learning-type super-resolution technique. Further, unless otherwise specified, the reconstruction constraint can be appropriately replaced with a process based on a dictionary.
- FIG. 8 is a block diagram showing an example of the configuration of an image processing apparatus 90 that restores a general super-resolution image using the technique described in Patent Document 1.
- the image processing apparatus 90 includes a reconstruction constraint calculation unit 910, a regularization term calculation unit 920, and an image restoration unit 930.
- the reconfiguration constraint calculation unit 910 calculates a reconfiguration constraint, that is, a reconfiguration cost.
- the regularization term calculation unit 920 calculates a constraint based on regularization (regularization term constraint), that is, a cost based on regularization.
- the regularization term calculation unit 920 of the image processing apparatus 90 using the super-resolution technique described in Patent Document 1 uses, for example, TV (Total Variation) as regularization (for example, see Non-Patent Document 1).
- TV is a method of regularizing so that the sum of absolute values of pixel value differences between adjacent pixels is minimized.
- the regularization term calculation unit 920 may use BTV (Bilateral Total Variation) (for example, see Non-Patent Document 2).
- BTV is a method of regularizing so that the sum of absolute values of pixel value differences between neighboring pixels is minimized, not limited to adjacent pixels.
- the BTV calculates the sum after multiplying the absolute value of the difference by a reduction coefficient based on the pixel position.
- the image restoration unit 930 generates (restores) a super-resolution image that minimizes the sum of the reconstruction cost and the regularization cost as the restored image.
- the image processing apparatus 90 using a general super-resolution technique restores a high-resolution image with an increased resolution in consideration of the reconstruction constraint and the regularization term constraint.
- the area included in the image can be classified into a plurality of types based on the feature of the pixel value.
- One of the classified areas is a texture area.
- the texture region is a region characterized in that a characteristic value of a predetermined pixel (for example, luminance gradation) is greatly different between adjacent or neighboring pixels. In such a texture region, it is desirable to store feature values (for example, gradation).
- regularization based on TV or BTV used by the image processing apparatus 90 averages (flattenes) feature values even for regions where feature values such as texture regions are to be stored. As a result, in the texture region, the feature value is flattened (for example, the luminance gradation is lowered).
- Patent Document 1 has a problem in that the feature value is flattened (for example, the gradation is reduced) in the region where the feature value (feature value) of the pixel is desired to remain.
- An object of the present invention is to solve the above-described problems and to provide an image processing apparatus and an image processing method for reducing flattening of an area where a pixel feature value is desired to be preserved in restoration of a super-resolution image.
- the image processing apparatus determines a region for storing the feature amount of the input image based on a gradient of the feature amount of the pixel of the input image and a direction of the gradient, and stores the feature amount.
- Weight calculation means for calculating a weight for reducing regularization term constraints, which are constraints based on regularization of image processing, and regularization term constraints for a high-resolution image restored based on the input image using the weights
- the reconstruction constraint calculation means for calculating the reconstruction constraint that is a constraint based on the reconstruction of the high-resolution image, the regularization term constraint, and the reconstruction constraint Image restoration means for restoring the high resolution image from the input image.
- An image processing method determines a region for storing a feature amount of the input image based on a gradient of a feature amount of a pixel of the input image and a direction of the gradient, and stores the feature amount.
- a recording medium that records a computer-readable program determines a region for storing a feature amount of the input image based on a gradient of the feature amount of a pixel of the input image and a direction of the gradient, A process for calculating a weight for reducing a regularization term constraint, which is a constraint based on regularization of image processing in the region for storing the feature quantity, and a high-resolution image restored based on the input image using the weight
- a program for causing a computer to execute processing for restoring the high-resolution image from the computer is stored.
- the present invention in the restoration of a super-resolution image, it is possible to reduce gradation reduction in an area where it is desired to leave a gradation.
- FIG. 1 is a block diagram showing an example of the configuration of the image processing apparatus according to the first embodiment of the present invention.
- FIG. 2 is a diagram for explaining an edge region.
- FIG. 3 is a diagram for explaining the texture region.
- FIG. 4 is a block diagram illustrating an example of the configuration of the weight calculation unit according to the first embodiment.
- FIG. 5 is a block diagram illustrating an example of another configuration of the image processing apparatus according to the first embodiment.
- FIG. 6 is a block diagram illustrating an example of the configuration of the image processing apparatus according to the second embodiment.
- FIG. 7 is a block diagram illustrating an example of the configuration of the image processing apparatus according to the third embodiment.
- FIG. 8 is a block diagram illustrating an example of a configuration of a general image processing apparatus.
- FIG. 1 is a block diagram showing an example of the configuration of the image processing apparatus 10 according to the first embodiment of the present invention.
- the image processing apparatus 10 includes a reconstruction constraint calculation unit 20, a regularization term calculation unit 30, an image restoration unit 40, and a weight calculation unit 50.
- the reconfiguration constraint calculation unit 20 calculates a value (reconfiguration constraint) indicating a constraint state based on the reconfiguration.
- the reconfiguration constraint calculation unit 20 of the present embodiment is not particularly limited as a value to be calculated (reconfiguration constraint).
- the reconstruction constraint calculation unit 20 may calculate a constraint as a reconstruction constraint so that the value (cost) increases when the constraint based on the reconstruction increases.
- the reconstruction constraint calculation unit 20 may calculate the value (cost) of the reconstruction constraint using Equation 1 below. good.
- the matrix D is a downsampling matrix
- the matrix B is a blurred matrix
- the matrix X is a matrix of an input image
- the matrix Y is a matrix of a super-resolution image (restored image).
- DBX is the product of the matrices D, B and X.
- the reconstruction constraint calculation unit 20 may calculate a constraint as a reconstruction constraint so that the value becomes smaller when the constraint based on the reconstruction becomes smaller.
- the regularization term calculation unit 30 calculates a value (regularization term constraint) indicating a constraint state based on regularization in consideration of the weight calculated by the weight calculation unit 50 described later.
- the regularization term calculation unit 30 is not particularly limited as a value to be calculated (regularization term constraint).
- the regularization term calculation unit 30 may calculate the constraint as a regularization term constraint so that the value decreases when the constraint is large.
- the regularization term calculation unit 30 may calculate the constraint so that the value becomes larger when the constraint becomes smaller as the regularization term constraint.
- the reconfiguration constraint calculation unit 20 and the regularization term calculation unit 30 of the present embodiment are not particularly limited in values to be calculated.
- each constraint will be described assuming that the value of each constraint increases when the constraint is large.
- the image processing apparatus 10 may replace the determination of the value in the following description.
- the image restoration unit 40 as a restored image, super-resolution that minimizes the sum (total cost) of the reconstruction constraint calculated by the reconstruction constraint calculation unit 20 and the regularization term constraint calculated by the regularization term calculation unit 30. Select (restore) an image.
- the image restoration unit 40 may select (restore) an image that minimizes the following Equation 2 as the super-resolution image.
- the first term of Formula 2 is the reconstruction constraint shown in Formula 1.
- the second term is a regularization term constraint.
- the image processing apparatus 10 repeats the operation described above until the total cost of the reconstruction constraint and the regularization term constraint is minimized in the restored image of the image restoration unit 40 or a predetermined number of times.
- the weight calculation unit 50 determines a region (for example, a texture region, which will be described below using a texture region as an example) that is characterized by a difference (for example, a gradation difference) between adjacent or neighboring pixels. Then, the weight calculation unit 50 calculates a weight such that a difference between pixels in the texture region remains, and sends the weight to the regularization term calculation unit 30.
- a region for example, a texture region, which will be described below using a texture region as an example
- a difference for example, a gradation difference
- the regularization term calculation unit 30 calculates a regularization term based on this weight.
- the weight calculation unit 50 determines a region in which the feature amount is stored, and calculates a weight for reducing the regularization term constraint in the region.
- the weight calculation unit 50 may reduce the constraint value in the texture region. Calculate the weight.
- the regularization term calculation unit 30 calculates a regularization term constraint (cost) in which a difference between pixels in the texture region is evaluated lower than a difference between pixels in other regions. That is, the regularization term calculation unit 30 calculates the regularization term constraint using a weight that is less affected by the difference between pixels in the texture region with respect to the regularization term constraint.
- the image processing apparatus 10 can restore an image in which a decrease (lower gradation) in a difference (for example, gradation) between pixels in the texture region is suppressed.
- the weight calculation unit 50 calculates the weight so that the constraint value becomes large in the texture region. Also good. Alternatively, the weight calculation unit 50 may calculate the weight in the same manner, and the regularization term calculation unit 30 may change the calculation method using the weight (for example, the product is inverted by sign inversion).
- the weight calculation unit 50 determines the texture area based on the feature value of the pixel in the texture area.
- the feature value of the pixel in the texture area is not particularly limited.
- the characteristic value of a pixel can be assumed to be a gradation of luminance and luminous intensity, a difference in color (hue), a difference in chroma (chroma), and the like.
- luminance as an example of the feature amount.
- FIG. 2 is a diagram for explaining an edge region including an end (edge).
- the edge region has a clear boundary (for example, a white and black boundary), and the boundary is generally in one direction.
- the edge region has a large amount of luminance gradient, and the luminance gradient direction is substantially uniform.
- the arrows of ⁇ 1 and ⁇ 2 shown in FIG. 2 are the main components of the luminance gradient (provided that ⁇ 1 ⁇ ⁇ 2 ). That is, the edge region has a large amount of luminance gradient, and thus “ ⁇ 1 > ⁇ 2 ”.
- the brightness gradient direction is aligned in the edge region.
- FIG. 3 is a diagram for explaining the texture region.
- ⁇ in FIG. 3 is the same as ⁇ in FIG. 2
- the texture region has a large amount of luminance gradient, but the direction of the luminance gradient is not uniform.
- the texture region has a large luminance gradient amount.
- the luminance gradient direction is not uniform in the texture region ( ⁇ 1 ⁇ 2 ).
- the texture area includes a forested forest area, an indoor carpet area, and a hair area on the head.
- the texture region is not limited to these. What kind of texture area is an area characterized by a feature value of a predetermined pixel (for example, luminance gradation) greatly different between adjacent or neighboring pixels in an image? It may be a simple area.
- the texture region of the present embodiment includes various regions including the above.
- the brightness gradient amount is small in the flat area.
- the luminance gradient direction may or may not be aligned. Note that there is no difference in luminance gradient components in a flat region, and the luminance gradient direction may not be obtained.
- the weight calculation unit 50 of the present embodiment detects a texture region based on the characteristics of each region described above, and calculates the weight of the detected texture region.
- FIG. 4 is a block diagram showing an example of the configuration of the weight calculation unit 50 of the present embodiment.
- the weight calculation unit 50 includes a gradient calculation unit 510, a direction calculation unit 520, and a weight data calculation unit 530.
- the gradient calculation unit 510 calculates a luminance gradient amount.
- the weight calculation unit 50 determines a flat region and other regions (edge region and texture region) based on the luminance gradient amount.
- the gradient calculation unit 510 of the present embodiment is not particularly limited as a method for calculating the luminance gradient amount, and various methods can be used.
- the gradient calculation unit 510 may calculate the luminance gradient amount as follows.
- the gradient calculation unit 510 calculates horizontal and vertical components (l x , ly ) of the luminance gradient amount of the entire image. Then, the gradient calculation unit 510 calculates a component of the luminance gradient amount using, for example, a Sobel filter (Sobel Filter).
- Sobel Filter Sobel Filter
- the gradient calculation unit 510 calculates a matrix (matrix I) whose diagonal component is the sum of squares of the components (l x , l y ) of the luminance gradient amount.
- the flat region has a small amount of luminance gradient, and the other regions (edge region and texture region) have a large amount of luminance gradient.
- the weight data calculation unit 530 which will be described later, can compare the sum of squares (diagonal component of the matrix I) with a predetermined threshold value to discriminate between a flat region and other regions (edge region and texture region).
- the gradient calculation unit 510 may store the sum of squares as a determination result.
- the gradient calculation unit 510 is not limited to the above, and the calculation result may be stored as storage information suitable for later processing.
- the gradient calculation unit 510 may store the determination result as a mask indicating a flat region.
- the gradient calculation unit 510 may create and store a mask (mask matrix M) in which a value “1” is set in a flat region and “0” is set in other regions.
- the direction calculation unit 520 calculates a luminance gradient direction amount. Then, the direction calculation unit 520 determines a texture region and an edge region based on the luminance gradient direction amount.
- the edge area has the same direction of brightness gradient.
- the brightness gradient direction is not aligned in the texture area.
- the direction calculation unit 520 operates as follows, for example.
- the direction calculation unit 520 calculates a pixel whose coordinates (x, y) satisfy the following Expression 3 as a straight pixel (edge).
- x and y are the horizontal and vertical coordinates of the pixel, respectively.
- Each variable is as follows.
- I x and I y luminance gradient components in the horizontal and vertical directions t 1 and t 2 : predetermined threshold values ⁇ 1 and ⁇ 2 : eigenvalues of the matrix A represented by the following equation 4, where ⁇ 1 > ⁇ 2 [Equation 4]
- R is a region including the pixel and neighboring pixels.
- the neighborhood is a predetermined range. For example, it is a range of three pixels in the front, rear, left, and right.
- the direction calculation unit 520 calculates the matrix J using the eigenvalues of the matrix A of the pixels satisfying Equation 3 as the degree to which the luminance gradient directions are aligned.
- the matrix J is a diagonal matrix.
- the diagonal component of the matrix J is a value calculated using the eigenvalue of the matrix A of each pixel that is satisfied by Equation 3.
- the diagonal component J i corresponding to the i-th pixel is expressed by the following Equation 5, where eigenvalues of the pixel (pixel i) are ⁇ i1 and ⁇ i2 .
- J 0 is stepped shape generated in the super-resolution image: Occurrence of (jaggy Jaggy), is a parameter determined in advance. Jaggy generated in the super-resolution image, as the value of J 0 is large, generation is suppressed.
- the image processing apparatus 10 in advance it is sufficient to store the J 0, there is no particular limitation on how to obtain J 0.
- the image processing apparatus 10 may read the J 0 from a database to save the unillustrated parameter may receive an input of J 0 from the user.
- ⁇ 0 is a predetermined constant.
- ⁇ a is a sigmoid function expressed by Expression 6 below.
- the weight data calculation unit 530 which will be described later, can determine the texture region and the edge region based on the matrix J calculated by the direction calculation unit 520.
- the weight data calculation unit 530 uses the regularization term based on the calculation result (luminance gradient amount) of the gradient calculation unit 510 and the calculation result of the direction calculation unit 520 (the degree to which the luminance gradient direction is aligned (matrix J)). The weight used by the calculation unit 30 is calculated. More specifically, the weight data calculation unit 530 determines a region including an edge region and a texture region based on the calculation result (luminance gradient amount) of the gradient calculation unit 510. Then, the weight data calculation unit 530 determines a texture region from the determined region using the result of the direction calculation unit 520. Then, the weight data calculation unit 530 calculates a weight obtained by reducing the weight of the determined texture region.
- the weight data calculation unit 530 transmits the calculated weight to the regularization term calculation unit 30.
- weight data calculation unit 530 does not have to limit the format of the weight data as long as the regularization term calculation unit 30 can use the weight.
- the weight data calculation unit 530 may transmit weights as vector data or matrix data.
- the weight data calculation unit 530 uses the calculation results of the gradient calculation unit 510 and the direction calculation unit 520 to calculate the weight represented by the following Expression 7.
- ⁇ is a vector
- diag ( ⁇ ) is a diagonal matrix (hereinafter referred to as “weight” or “weight matrix”) in which the components of the vector ⁇ are arranged in a diagonal component.
- ⁇ and ⁇ are predetermined constants.
- the matrix I and the matrix M are mask matrices calculated from the unit matrix and the square sum matrix calculated by the gradient calculation unit 510.
- the matrix J is a calculation result indicating the degree to which the luminance gradients calculated by the direction calculation unit 520 are aligned.
- the weight data calculation unit 530 transmits the calculated weight to the regularization term calculation unit 30.
- the regularization term calculation unit 30 that has received this weight calculates a regularization term constraint using the weight.
- the regularization term calculation unit 30 uses the regularization term constraint R (X) shown in the following Equation 8.
- the matrix X is an input image.
- the matrix S is a matrix indicating parallel movement.
- the subscript of S indicates the direction and the number of pixels that the superscript moves.
- S l x is a matrix that translates l pixels in the x direction.
- the constraint R (X) of the regularization term calculation unit 30 is not limited to Equation 8.
- the regularization term calculation unit 30 may use the constraint R (X) shown in the following formula 9 instead of the formula 8.
- p and q are parameters determined by the user.
- each configuration of the image processing apparatus 10 has been described as sending data to the next configuration.
- the image processing apparatus 10 need not be limited to this configuration.
- the image processing apparatus 10 may include a storage unit (not shown), and each configuration may store a result in the storage unit and retrieve necessary data from the storage unit.
- ⁇ and ⁇ are set as predetermined constants, and p and q are set as predetermined parameters.
- the present embodiment is not limited to this.
- the image processing apparatus 10 may determine ⁇ , ⁇ , p, and q based on a predetermined method (for example, regression analysis) in advance for each pixel or for each region.
- the image processing apparatus 10 artificially generates a learning high resolution image (hereinafter referred to as “X H ”) prepared in advance and the high resolution image (X H ).
- X H learning high resolution image
- ⁇ , ⁇ , p, and q may be obtained based on the generated low-resolution image (hereinafter referred to as “Y L ”).
- the image processing apparatus 10 uses ⁇ , ⁇ , and so on so that the function R (X H ) expressed by Equation 9 is minimized for each pixel or region of the plurality of high resolution images (X H ) for learning. Find the set of p and q. Then, the image processing apparatus 10 calculates a function for calculating ⁇ , ⁇ , p, and q from the value of the low resolution image (Y L ) generated from the high resolution image (X H ) for each pixel or each region. What is necessary is just to obtain
- the function of the image processing apparatus 10 is, for example, various J L calculated by applying Formula 5 to the low resolution image (Y L ), and ⁇ , ⁇ , p, and J corresponding to J L. Based on q, ⁇ , ⁇ , p, and q functions with J L as an argument are calculated. (Hereinafter, the respective functions are referred to as ⁇ L (J L ), ⁇ L (J L ), p L (J L ), and q L (J L )) or more directly, the image processing apparatus 10. Holds various J L and corresponding ⁇ , ⁇ , p, and q on the lookup table, and refers to the lookup table to determine the values of these ( ⁇ , ⁇ , p, and q). You may ask.
- the image processing apparatus 10 also calculates J L using Equation 5 for the low-resolution input image (Y L ) that is the target of super-resolution imaging. Then, the image processing apparatus 10, based on the calculated J L, the above function ( ⁇ L (J L), ⁇ L (J L), p L (J L), and q L (J L)) Alternatively, ⁇ , ⁇ , p, and q may be calculated using a lookup table.
- the calculation method of ⁇ , ⁇ , p, and q of the image processing apparatus 10 is not limited to this.
- the image processing apparatus 10 may use the following method.
- the image processing apparatus 10 has the minimum J H calculated by applying Formula 5 to the high resolution image (X H ) and the function R (X H ) expressed by Formula 9 in advance for each pixel or area. A set of ⁇ , ⁇ , p, and q is calculated. Then, the image processing apparatus 10 calculates the relationship between J H and ⁇ , ⁇ , p, and q based on a predetermined method (for example, regression analysis) in the same manner as described above.
- a predetermined method for example, regression analysis
- the image processing apparatus 10 calculates J L by applying Equation 5 to the input low resolution image (Y L ). Thereafter, the image processing apparatus 10 uses the above-described functions ( ⁇ L (J L ), ⁇ L (J L ), p L (J L ), and q L (J L )) or a look-up table to determine ⁇ , ⁇ , p, and q are provisionally calculated. Then, the image processing apparatus 10 generates a super-resolution image (hereinafter referred to as “X SR ”) using these ( ⁇ , ⁇ , p, and q).
- X SR super-resolution image
- the image processing apparatus 10 calculates JSR by applying Equation 5 to the super-resolution image (X SR ).
- the image processing apparatus 10 determines that ⁇ is based on JSR and the above functions ( ⁇ H (J H ), ⁇ H (J H ), p H (J H ), and q H (J H )). , ⁇ , p, and q are calculated again. Then, the image processing apparatus 10 updates the super-resolution image (X SR ) using the calculated ⁇ , ⁇ , p, and q.
- the image processing apparatus 10 repeats the above processing until there is no change in the values of ⁇ , ⁇ , p, and q from the super-resolution image (X SR ), or the changes are within a predetermined range. As described above, the image processing apparatus 10 updates the super-resolution image (X SR ).
- the image processing apparatus 10 may calculate ⁇ , ⁇ , p, and q as follows.
- the image processing apparatus 10 applies Formula 5 to J L calculated by applying Formula 5 to the input low-resolution image (Y L ) or Super-resolution image (X SR ). using the J SR which is calculated by, alpha, beta, were calculated p, and q. However, the image processing apparatus 10 may calculate ⁇ , ⁇ , p, and q using both J L and J SR .
- the image processing apparatus 10 in advance, for each pixel or for each region, and J H calculated by applying the formula 5 in the high resolution image (X H), low generated from the high resolution image (X H) J L calculated by applying Formula 5 to the resolution image (Y L ) is calculated. Furthermore, the image processing apparatus 10 calculates ⁇ , ⁇ , p, and q that minimize the function R (X H ) expressed by Equation 9.
- the image processing apparatus 10 uses J L and J H and ⁇ , ⁇ , p, and q corresponding to each of them in a predetermined method (for example, regression analysis), and uses J L and J H as arguments.
- a predetermined method for example, regression analysis
- J L and J H uses J L and J H as arguments.
- ⁇ LH (J L , J H ) ⁇ LH (J L , J H ), ⁇ LH (J L , J H ), p LH (J L , J H ), and q LH (J L , J H )
- the image processing apparatus 10 holds the relationship between “ ⁇ , ⁇ , p, and q” and “J L and J H ” as a lookup table.
- the image processing apparatus 10 obtains J L and J SR based on the super-resolution image (X SR ) and the input image (Y L ) generated using the same method as described above. Then, the image processing apparatus 10 calculates ⁇ , ⁇ , p, and q based on J L and J SR . Then, the image processing apparatus 10 updates the super-resolution image (X SR ) based on the calculated ⁇ , ⁇ , p, and q. Then, the image processing apparatus 10 calculates J L and J SR based on the updated super-resolution image (X SR ) and the input image (Y L ).
- the image processing apparatus 10 then calculates ⁇ LH (J L , J H ), ⁇ LH (J L , J H ), p LH (J L , J H ), based on the calculated J L and J SR . And q LH (J L , J H ) are used to update ⁇ , ⁇ , p and q.
- the image processing apparatus 10 repeats this process until the values of the super-resolution image (X SR ), ⁇ , ⁇ , p, and q do not change or the value changes fall within a predetermined range.
- the image processing apparatus 10 can obtain the effect of reducing the gradation reduction of the texture region where the gradation is desired to be retained in the restoration of the high resolution image.
- the weight calculation unit 50 of the image processing apparatus 10 determines the texture region based on the magnitude of the luminance gradient amount and the degree to which the luminance gradient amount is uniform. Then, the weight calculation unit 50 calculates the weight so that the difference between pixels is large and the calculation result of the regularization term (regularization term constraint) is low for the determined texture region. And the regularization term calculation part 30 calculates a regularization term using the weight. Therefore, the image restoration unit 40 can select (restore) a restored image that does not reduce the gradation of the texture region.
- the configuration of the image processing apparatus 10 is not limited to the above description.
- the image processing apparatus 10 may include a storage unit (not shown) that has already been described.
- the image processing apparatus 10 may divide each configuration into a plurality of configurations.
- the weight calculation unit 50 may divide the direction calculation unit 520 into a matrix A calculation unit, an eigenvalue calculation unit, and a matrix J calculation unit.
- the image processing apparatus 10 may have a plurality of configurations as one configuration.
- the weight calculation unit 50 and the regularization term calculation unit 30 may have one configuration.
- the image processing apparatus 10 of the present embodiment may be realized as a computer including a CPU (Central Processing Unit), a ROM (Read Only Memory), and a RAM (Random Access Memory). Further, the image processing apparatus 10 includes an IO (Input / Output Unit) and a NIC (Network Interface Circuit or Network interface Card), and may be connected to other devices or devices.
- a CPU Central Processing Unit
- ROM Read Only Memory
- RAM Random Access Memory
- the image processing apparatus 10 includes an IO (Input / Output Unit) and a NIC (Network Interface Circuit or Network interface Card), and may be connected to other devices or devices.
- IO Input / Output Unit
- NIC Network Interface Circuit or Network interface Card
- FIG. 5 is a block diagram showing an example of the configuration of the image processing apparatus 18 which is another configuration of the present embodiment.
- the image processing device 18 includes a CPU 810, a ROM 820, a RAM 830, an internal storage device 840, an IO 850, an input device 860, a display device 870, and a NIC 880, and constitutes a computer.
- CPU 810 reads the program from ROM 820 or internal storage device 840. Then, based on the read program, the CPU 810 performs the reconstruction constraint calculation unit 20, the regularization term calculation unit 30, the image restoration unit 40, and the weight calculation unit 50 of the image processing apparatus 10 illustrated in FIG. Realize the function.
- the CPU 810 uses the RAM 830 or the internal storage device 840 as temporary storage when realizing each function.
- the CPU 810 receives input data from the input device 860 via the IO 850 and outputs the data to the display device 870.
- the CPU 810 may operate by reading a program included in the storage medium 890 storing the program so as to be readable by a computer into the RAM 830 using a storage medium reading device (not shown). As described above, the CPU 810 may use a non-transitory memory such as the ROM 810 or the storage medium 890, or may use a volatile memory such as the RAM 830.
- the CPU 810 may receive a program from an external device (not shown) via the NIC 880.
- ROM 820 stores a program executed by CPU 810 and fixed data.
- the ROM 820 is, for example, a P-ROM (Programmable-ROM) or a flash ROM.
- the RAM 830 temporarily stores programs and data executed by the CPU 810.
- the RAM 830 is, for example, a D-RAM (Dynamic-RAM).
- the internal storage device 840 stores data and programs that the image processing device 18 stores for a long time. Further, the internal storage device 840 may operate as a temporary storage device for the CPU 810.
- the internal storage device 840 is, for example, a hard disk device, a magneto-optical disk device, an SSD (Solid State Drive), or a disk array device.
- the IO 850 mediates data between the CPU 810, the input device 860, and the display device 870.
- the IO850 is, for example, an IO interface card or a USB (Universal Serial Bus) card.
- the input device 860 is an input unit that receives an input instruction from an operator of the image processing apparatus 18.
- the input device 860 is, for example, a keyboard, a mouse, or a touch panel.
- the display device 870 is a display unit of the image processing apparatus 18.
- the display device 870 is a liquid crystal display, for example.
- the NIC 880 relays exchange of data (images) with other devices (for example, a device that transmits an input image (not shown) and a device that receives a restored image) via a network.
- the NIC 880 is, for example, a LAN (Local Area Network) card.
- the image processing apparatus 18 configured as described above can obtain the same effects as the image processing apparatus 10.
- the edges included in the edge region are often smooth. However, the edge may be stepped (jaggy).
- the determination of whether an edge is smooth or jaggy requires reference to a plurality of pixels. However, since TV and BTV use differences between individual pixels, it is impossible to distinguish whether the edge is smooth or jaggy. Therefore, the technique described in Patent Document 1 has a problem that jaggy noise is generated in an edge region including a smooth edge.
- the image processing apparatus 11 suppresses the occurrence of jaggy noise in the edge region.
- FIG. 6 is a block diagram illustrating an example of the configuration of the image processing apparatus 11 according to the second embodiment.
- the image processing apparatus 11 includes a reconstruction constraint calculation unit 20, a regularization term calculation unit 30, an image restoration unit 41, a weight calculation unit 51, and a direction constraint calculation unit 60.
- the image processing apparatus 11 of the present embodiment may be realized by a computer including the CPU 810, the ROM 820, and the RAM 830 shown in FIG. 5.
- the weight calculation unit 51 operates in the same manner as the weight calculation unit 50 of the first embodiment, except that the calculated weight is sent to the regularization term calculation unit 30 and is also sent to the direction constraint calculation unit 60. Therefore, detailed description other than this is omitted.
- the direction constraint calculation unit 60 smoothes the pixel value in the edge direction for a region where the direction (edge direction) orthogonal to the maximum luminance gradient direction is aligned, such as an edge included in the edge region. Constraint (direction constraint) is calculated.
- the direction constraint calculation unit 60 may use a differential value in the edge direction as a value indicating the smoothness used for calculating the direction constraint.
- the direction constraint calculation unit 60 may calculate the direction constraint as follows.
- the direction constraint calculation unit 60 imposes a small penalty on a region where the change in the differential value is small as a region where the edge direction is smooth (reducing the direction constraint). On the other hand, the direction constraint calculation unit 60 imposes a large penalty on a region where the change in the differential value is large as a region where the edge direction is not smooth (increases the direction constraint).
- the direction constraint calculation part 60 should just calculate the direction constraint represented by the following Numerical formula 10, for example.
- Equation 11 the matrix L n is a differentiation along the edge direction represented by the following equation (Equation 11).
- ⁇ is an angle between the horizontal direction (x-axis) and the edge direction.
- the image restoration unit 41 restores the image in consideration of the direction constraint calculated by the direction constraint calculation unit 60 in addition to the reconstruction constraint and the regularization term constraint.
- the image restoration unit 41 selects (restores) an image that minimizes the following Expression 12.
- Equation 12 The third term of Equation 12 is a direction constraint.
- the first and second terms are the same as in Equation 2.
- the image processing apparatus 11 can obtain an effect of suppressing the occurrence of jaggy noise in the edge region.
- the direction constraint calculation unit 60 of the image processing apparatus 11 calculates a direction constraint that applies a penalty larger than the smooth region to the region that is not smooth in the edge direction. This is because the image restoration unit 41 restores an image in consideration of the direction constraint, and thus can restore an image that leaves a region with a smooth edge direction.
- the image processing apparatus 12 according to the third embodiment can reduce the processing time.
- FIG. 7 is a block diagram illustrating an example of the configuration of the image processing apparatus 12 according to the third embodiment.
- the image processing apparatus 12 includes a reconstruction constraint calculation unit 20, a regularization term calculation unit 31, an image restoration unit 40, a weight calculation unit 50, and a reference pixel pruning unit 70.
- image processing device 12 of the third embodiment may include the direction constraint calculation unit 60 of the second embodiment.
- the image processing apparatus 12 of the present embodiment may be realized by a computer including the CPU 810, the ROM 820, and the RAM 830 shown in FIG. 5.
- the reconstruction constraint calculation unit 20 the image restoration unit 40, and the weight calculation unit 50 are the same as those of the image processing apparatus 10 of the first embodiment, detailed description thereof is omitted.
- the reference pixel pruning unit 70 performs processing so that a part of the weight calculated by the weight calculation unit 50 is not used for calculation of the regularization term constraint in the regularization term calculation unit 31. This process is hereinafter referred to as “pruning”.
- the regularization term calculation unit 31 calculates a regularization term using the weights pruned by the reference pixel pruning unit 70. Therefore, the regularization term calculation unit 31 can reduce the number of pixels used for calculation of the regularization term.
- the pruning process of the reference pixel pruning unit 70 of the present embodiment is not particularly limited.
- the reference pixel pruning unit 70 may include a mask corresponding to a predetermined pixel and apply the mask to the weight.
- the regularization term calculation unit 31 does not calculate the regularization term constraint for the pixel corresponding to the masked weight.
- the reference pixel pruning unit 70 may randomly select a weight component and set the value of the component to “0”. For example, when the weight calculation unit 50 calculates the weight shown in Equation 7, the reference pixel pruning unit 70 randomly selects a component of the vector ⁇ or the weight matrix (diag ( ⁇ )), The value may be “0”.
- the regularization term calculation unit 31 does not calculate the difference between the components having the value “0”.
- the regularization term calculation unit 30 reduces the difference calculation process based on the result of the reference pixel pruning unit 70.
- the reference pixel pruning unit 70 may notify the regularization term calculation unit 31 of pruning information (for example, a flag) as information different from the weight without operating the weight.
- the regularization term calculation unit 31 receives the weight from the weight calculation unit 50 and calculates the regularization term constraint based on the pruning information from the reference pixel pruning unit 70 as in the first embodiment. What is necessary is just to reduce the pixel to perform.
- the image processing apparatus 12 can obtain an effect of reducing the processing time.
- the reference pixel pruning unit 70 of the image processing device 12 sets a reduction of pixels for which no difference is calculated. As a result, the regularization term calculation unit 31 can reduce the difference calculation process.
- saves the feature-value of the said input image is determined based on the gradient of the feature-value of the pixel of an input image, and the direction of the said gradient, and regularization of the image processing in the area
- An image processing apparatus including image restoration means for restoring an image.
- the weight calculation means includes a gradient calculation means for calculating the gradient of the feature amount of the pixel of the input image, a direction calculation means for calculating the degree of alignment of the gradient direction of the feature amount, and the gradient And weight data calculating means for calculating the weight based on the degree of alignment of the directions.
- the weight data calculation means calculates the weight reduced by the predetermined value in a region where the magnitude of the gradient is larger than a predetermined value and the degree of alignment of the gradient is smaller than the predetermined value.
- the said regularization term calculation means is based on the high resolution image for learning, the low resolution image artificially generated from the high resolution image, and the weight calculated from the weight calculation means. 4.
- the image processing apparatus according to any one of appendix 1 to appendix 3, wherein a constant necessary for expressing the regularization term is calculated.
- the regularization term calculation means includes a high-resolution image for learning, a low-resolution image artificially generated from the high-resolution image, a super-resolution image temporarily generated,
- the image processing apparatus according to any one of appendix 1 to appendix 4, wherein a constant necessary for expressing the regularization term is calculated by repetition processing from the weight calculated by the weight calculation means. .
- limiting and regularization are included including the direction restriction
- the image processing apparatus according to any one of appendices 1 to 5, wherein the image is restored using the direction constraint in addition to the term constraint.
- supplementary note 7 In any one of supplementary notes 1 to 6, further comprising reference pixel pruning means for restricting a part of the pixels from being used in the calculation of the regularization term constraint of the regularization term calculation means.
- Image processing device In any one of supplementary notes 1 to 6, further comprising reference pixel pruning means for restricting a part of the pixels from being used in the calculation of the regularization term constraint of the regularization term calculation means.
- Appendix 8 The image processing apparatus according to appendix 7, wherein the reference pixel pruning unit deletes a part of the weight calculated by the weight calculation unit.
- saves the feature-value of the said input image based on the gradient of the feature-value of the pixel of the input image and the said gradient is determined, and the image processing in the area
- the weight which reduced the said predetermined value is calculated in the area
- saves the feature-value of the said input image based on the gradient of the feature-value of the pixel of the input image and the said gradient is determined, and the image processing in the area
- a computer-readable recording medium storing a program to be executed.
- a high-resolution image for learning, a low-resolution image artificially generated from the high-resolution image, a super-resolution image temporarily generated, and the weight calculated by the weight calculation unit A computer-readable recording medium storing the program according to any one of appendix 17 to appendix 20, which causes a computer to execute a process of calculating a constant necessary for expressing the regularization term by an iterative process .
- Supplementary Note 17 to Supplementary Note 21 that causes a computer to execute processing for restoring an image is a computer-readable recording medium recording the program according to Item 1.
- weight calculation means for calculating a weight for reducing a regularization term constraint that is a constraint based on regularization of image processing, and the texture area uses the weight.
- the regularization term calculation means for calculating the regularization term constraint without using the weight, the regularization term constraint, and the reconstruction that is a constraint based on the reconstruction of the high-resolution image
- An image processing apparatus comprising: a constraint; and an image restoration unit that restores a high-resolution image based on the constraint.
- a weight for reducing the regularization term constraint which is a constraint based on regularization of image processing, is calculated, the weight is used in the texture region, and the region other than the texture region is used. Calculate the regularization term constraint without using the weight, and based on the regularization term constraint and a reconstruction constraint that is a constraint based on reconstruction of the high-resolution image Image processing method to restore.
Landscapes
- Engineering & Computer Science (AREA)
- Multimedia (AREA)
- Signal Processing (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Image Processing (AREA)
- Editing Of Facsimile Originals (AREA)
Abstract
Description
図1は、本発明における第1の実施形態に係る画像処理装置10の構成の一例を示すブロック図である。
テクスチャ領域は、図3に示すように、輝度勾配量が大きいが、輝度勾配の方向が揃っていない。(例えば、図3には、水平方向の境界と、垂直方向の境界がある。)つまり、テクスチャ領域は、輝度勾配量が大きい。また、テクスチャ領域は、輝度勾配方向が揃っていない(λ1≒λ2)。
さらに、方向算出部520は、数式3を満たす画素の行列Aの固有値を用いて、輝度勾配方向の揃っている程度として、行列Jを算出する。
画像処理装置10の構成は、これまでの説明に限らない。
エッジ領域に含まれるエッジは、滑らかな場合が多い。ただし、エッジは、階段状(ジャギー:jaggy)となっている場合もある。エッジが滑らか又はジャギーであるかの判定は、複数の画素の参照が必要である。しかし、TV及びBTVは、個々の画素間の差分を用いるため、エッジが滑らかであるかジャギーであるかを区別できない。そのため、特許文献1に記載の技術は、滑らかなエッジを含むエッジ領域にジャギーノイズが発生する問題点があった。
TV及びBTVは、対象範囲の全て画素間の差分を算出することが必要である。特に、BTVは、近傍領域の全ての画素間の差分を算出することが必要である。そのため、特許文献1に記載の技術は、大きな処理時間を必要とする。
を含む画像処理装置。
11 画像処理装置
12 画像処理装置
18 画像処理装置
20 再構成制約算出部
30 正則化項算出部
31 正則化項算出部
40 画像復元部
41 画像復元部
50 重み算出部
51 重み算出部
60 方向制約算出部
70 参照画素枝刈り部
90 画像処理装置
510 勾配算出部
520 方向算出部
530 重みデータ算出部
810 CPU
820 ROM
830 RAM
840 内部記憶装置
850 IO
860 入力機器
870 表示機器
880 NIC
890 記憶媒体
910 再構成制約算出部
920 正則化項算出部
930 画像復元部
Claims (11)
- 入力画像の画素の特徴量の勾配と前記勾配の方向とを基に前記入力画像の特徴量を保存する領域を判定し、前記特徴量を保存する領域における画像処理の正則化に基づく制約である正則化項制約を低減する重みを算出する重み算出手段と、
前記重みを用いて前記入力画像を基に復元される高解像画像の正則化項制約を算出する正則化項算出手段と、
前記高解像画像の再構成に基づく制約である再構成制約を算出する再構成制約算出手段と、
前記正則化項制約と前記再構成制約とを基に前記入力画像から前記高解像画像を復元する画像復元手段と
を含む画像処理装置。 - 前記重み算出手段は、
前記入力画像の画素の特徴量の勾配を算出する勾配算出手段と、
前記特徴量の勾配の方向の揃っている程度を算出する方向算出手段と、
前記勾配と前記方向の揃っている程度を基に前記重みを算出する重みデータ算出手段と
を含む請求項1に記載の画像処理装置。 - 前記重みデータ算出手段が、
前記勾配の大きさが所定の値より大きく、
前記勾配の方向の揃っている程度が所定の値より値小さい領域において
前記所定の値低減した重みを算出する
ことを特徴とする請求項1又は請求項2に記載の画像処理装置。 - 前記復元画像において前記入力画像の特徴量の勾配方向に直行するエッジ方向が滑らかとなる方向制約を算出する方向制約算出手段を含み、
前記画像復元手段が、再構成制約及び正則化項制約に加え、前記方向制約を用いて画像を復元する
ことを特徴とする請求項1乃至請求項3のいずれが1項に記載の画像処理装置。 - 前記正則化項算出手段の正則化項制約の算出において一部の画素を使用しないように制限する参照画素枝刈り手段を
さらに含む請求項1乃至請求項4のいずれか1項に記載に画像処理装置。 - 前記参照画素枝刈り手段は、
前記重み算出手段が算出する重みの一部を削除する
ことを特徴とする請求項5に記載の画像処理装置。 - 入力画像の画素の特徴量の勾配と前記勾配の方向とを基に前記入力画像の特徴量を保存する領域を判定し、前記特徴量を保存する領域における画像処理の正則化に基づく制約である正則化項制約を低減する重みを算出し、
前記重みを用いて前記入力画像を基に復元される高解像画像の正則化項制約を算出し、
前記高解像画像の再構成に基づく制約である再構成制約を算出し、
前記正則化項制約と前記再構成制約とを基に前記入力画像から前記高解像画像を復元する
画像処理方法。 - 入力画像の画素の特徴量の勾配と前記勾配の方向とを基に前記入力画像の特徴量を保存する領域を判定し、前記特徴量を保存する領域における画像処理の正則化に基づく制約である正則化項制約を低減する重みを算出する処理と、
前記重みを用いて前記入力画像を基に復元される高解像画像の正則化項制約を算出する処理と、
前記高解像画像の再構成に基づく制約である再構成制約を算出する処理と、
前記正則化項制約と前記再構成制約とを基に前記入力画像から前記高解像画像を復元する処理と
をコンピュータに実行させるプログラムを記録したコンピュータ読み取り可能な記録媒体。 - 入力画像のテクスチャ領域において、画像処理の正則化に基づく制約である正則化項制約を低減する重みを算出する重み算出手段と、
前記テクスチャ領域においては前記重みを用い、前記テクスチャ領域以外の領域においては、前記重みを用いることなく、前記正則化項制約を算出する正則化項算出手段と、
前記正則化項制約と、前記高解像画像の再構成に基づく制約である再構成制約と、に基づいて高解像画像を復元する画像復元手段と
を含む画像処理装置。 - 入力画像のテクスチャ領域において、画像処理の正則化に基づく制約である正則化項制約を低減する重みを算出し、
前記テクスチャ領域においては前記重みを用い、前記テクスチャ領域以外の領域においては、前記重みを用いることなく、前記正則化項制約を算出し、
前記正則化項制約と、前記高解像画像の再構成に基づく制約である再構成制約と、に基づいて高解像画像を復元する
画像処理方法。 - 入力画像のテクスチャ領域において、画像処理の正則化に基づく制約である正則化項制約を低減する重みを算出する処理と、
前記テクスチャ領域においては前記重みを用い、前記テクスチャ領域以外の領域においては、前記重みを用いることなく、前記正則化項制約を算出する処理と、
前記正則化項制約と、前記高解像画像の再構成に基づく制約である再構成制約と、に基づいて高解像画像を復元する処理と
をコンピュータに実行させるプログラムを記録したコンピュータ読み取り可能な記録媒体。
Priority Applications (4)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
EP13857754.9A EP2927864B1 (en) | 2012-11-29 | 2013-11-28 | Image processing device and image processing method |
JP2014549833A JP6443050B2 (ja) | 2012-11-29 | 2013-11-28 | 画像処理装置、及び、画像処理方法 |
US14/647,531 US9542725B2 (en) | 2012-11-29 | 2013-11-28 | Image processing device, image processing method and medium |
HK15111799.3A HK1211125A1 (en) | 2012-11-29 | 2015-12-01 | Image processing device and image processing method |
Applications Claiming Priority (2)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
JP2012260844 | 2012-11-29 | ||
JP2012-260844 | 2012-11-29 |
Publications (1)
Publication Number | Publication Date |
---|---|
WO2014083857A1 true WO2014083857A1 (ja) | 2014-06-05 |
Family
ID=50827517
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
PCT/JP2013/007001 WO2014083857A1 (ja) | 2012-11-29 | 2013-11-28 | 画像処理装置、及び、画像処理方法 |
Country Status (5)
Country | Link |
---|---|
US (1) | US9542725B2 (ja) |
EP (1) | EP2927864B1 (ja) |
JP (1) | JP6443050B2 (ja) |
HK (1) | HK1211125A1 (ja) |
WO (1) | WO2014083857A1 (ja) |
Cited By (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
WO2020179205A1 (ja) * | 2019-03-01 | 2020-09-10 | 株式会社ソニー・インタラクティブエンタテインメント | 画像送受信システム、画像送信装置、画像受信装置、画像送受信方法及びプログラム |
Families Citing this family (10)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US10410398B2 (en) * | 2015-02-20 | 2019-09-10 | Qualcomm Incorporated | Systems and methods for reducing memory bandwidth using low quality tiles |
CN106169174B (zh) * | 2016-06-30 | 2019-12-31 | 北京大学 | 一种图像放大方法 |
CN106169173B (zh) * | 2016-06-30 | 2019-12-31 | 北京大学 | 一种图像插值方法 |
CN106447614B (zh) * | 2016-11-25 | 2019-06-07 | 浙江师范大学 | 一种基于一致性区域的超分辨率方法 |
CN107862655B (zh) * | 2017-10-26 | 2020-02-18 | 电子科技大学 | 一种基于正则化的交替最小化高分辨率图像重构方法 |
CN109741254B (zh) * | 2018-12-12 | 2022-09-27 | 深圳先进技术研究院 | 字典训练及图像超分辨重建方法、***、设备及存储介质 |
US11037339B2 (en) * | 2019-04-17 | 2021-06-15 | Uih America, Inc. | Systems and methods for image reconstruction |
CN110211036A (zh) * | 2019-04-25 | 2019-09-06 | 中国人民解放军61646部队 | 一种基于智能先验的图像超分辨率重建方法及装置 |
TWI703872B (zh) | 2019-05-15 | 2020-09-01 | 瑞昱半導體股份有限公司 | 影像色彩還原與增強電路 |
CN117541495A (zh) * | 2023-09-04 | 2024-02-09 | 长春理工大学 | 一种自动优化模型权重的图像条纹去除方法、装置及介质 |
Citations (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
WO2004068862A1 (ja) * | 2003-01-31 | 2004-08-12 | The Circle For The Promotion Of Science And Engineering | 高解像度カラー画像生成方法,高解像度カラー画像生成装置及び高解像度カラー画像生成プログラム |
US20050270386A1 (en) * | 2004-05-28 | 2005-12-08 | Hirofumi Saitoh | Method and apparatus for authentication utilizing iris |
JP2008167027A (ja) * | 2006-12-27 | 2008-07-17 | Olympus Corp | 画像処理装置、画像処理方法、画像処理プログラム |
JP2011180798A (ja) | 2010-03-01 | 2011-09-15 | Sony Corp | 画像処理装置、および画像処理方法、並びにプログラム |
Family Cites Families (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
GB0126065D0 (en) * | 2001-10-31 | 2001-12-19 | Elam T Ltd | Document authentication |
US9117133B2 (en) * | 2008-06-18 | 2015-08-25 | Spectral Image, Inc. | Systems and methods for hyperspectral imaging |
US8913831B2 (en) * | 2008-07-31 | 2014-12-16 | Hewlett-Packard Development Company, L.P. | Perceptual segmentation of images |
JP5506272B2 (ja) * | 2009-07-31 | 2014-05-28 | 富士フイルム株式会社 | 画像処理装置及び方法、データ処理装置及び方法、並びにプログラム |
US8180146B2 (en) * | 2009-12-22 | 2012-05-15 | The Chinese University Of Hong Kong | Method and apparatus for recognizing and localizing landmarks from an image onto a map |
-
2013
- 2013-11-28 US US14/647,531 patent/US9542725B2/en active Active
- 2013-11-28 EP EP13857754.9A patent/EP2927864B1/en active Active
- 2013-11-28 WO PCT/JP2013/007001 patent/WO2014083857A1/ja active Application Filing
- 2013-11-28 JP JP2014549833A patent/JP6443050B2/ja active Active
-
2015
- 2015-12-01 HK HK15111799.3A patent/HK1211125A1/xx unknown
Patent Citations (7)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
WO2004068862A1 (ja) * | 2003-01-31 | 2004-08-12 | The Circle For The Promotion Of Science And Engineering | 高解像度カラー画像生成方法,高解像度カラー画像生成装置及び高解像度カラー画像生成プログラム |
AU2003272936A1 (en) * | 2003-01-31 | 2004-08-23 | The Circle For The Promotion Of Science And Engineering | Method for creating high resolution color image, system for creating high resolution color image and program for creating high resolution color image |
US20060038891A1 (en) * | 2003-01-31 | 2006-02-23 | Masatoshi Okutomi | Method for creating high resolution color image, system for creating high resolution color image and program creating high resolution color image |
US20050270386A1 (en) * | 2004-05-28 | 2005-12-08 | Hirofumi Saitoh | Method and apparatus for authentication utilizing iris |
JP2005334402A (ja) * | 2004-05-28 | 2005-12-08 | Sanyo Electric Co Ltd | 認証方法および認証装置 |
JP2008167027A (ja) * | 2006-12-27 | 2008-07-17 | Olympus Corp | 画像処理装置、画像処理方法、画像処理プログラム |
JP2011180798A (ja) | 2010-03-01 | 2011-09-15 | Sony Corp | 画像処理装置、および画像処理方法、並びにプログラム |
Non-Patent Citations (2)
Title |
---|
FARSIU, S.; ROBINSON, M.D.; ELAD, M.; MILANFAR, P.: "Fast and robust multiframe super resolution", IEEE TRANSACTIONS ON IMAGE PROCESSING, vol. 13, no. 10, 2004, pages 1327 - 1344 |
MICHAEL K. NG; HUANFENG SHEN; EDMUND Y. LAM; LIANGPEI ZHANG: "A Total Variation Regularization Based Super-Resolution Reconstruction Algorithm for Digital Video", EURASIP JOURNAL ON ADVANCES IN SIGNAL PROCESSING, vol. 2007, pages 16 |
Cited By (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
WO2020179205A1 (ja) * | 2019-03-01 | 2020-09-10 | 株式会社ソニー・インタラクティブエンタテインメント | 画像送受信システム、画像送信装置、画像受信装置、画像送受信方法及びプログラム |
JPWO2020179205A1 (ja) * | 2019-03-01 | 2021-11-11 | 株式会社ソニー・インタラクティブエンタテインメント | 画像送受信システム、画像送信装置、画像受信装置、画像送受信方法及びプログラム |
JP7116244B2 (ja) | 2019-03-01 | 2022-08-09 | 株式会社ソニー・インタラクティブエンタテインメント | 画像送受信システム、画像送信装置、画像受信装置、画像送受信方法及びプログラム |
Also Published As
Publication number | Publication date |
---|---|
US9542725B2 (en) | 2017-01-10 |
EP2927864B1 (en) | 2018-06-13 |
EP2927864A4 (en) | 2016-08-17 |
JPWO2014083857A1 (ja) | 2017-01-05 |
HK1211125A1 (en) | 2016-05-13 |
US20150317767A1 (en) | 2015-11-05 |
JP6443050B2 (ja) | 2018-12-26 |
EP2927864A1 (en) | 2015-10-07 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
JP6443050B2 (ja) | 画像処理装置、及び、画像処理方法 | |
US8675999B1 (en) | Apparatus, system, and method for multi-patch based super-resolution from an image | |
JP6362333B2 (ja) | 画像処理装置、画像処理方法、およびプログラム | |
US7983511B1 (en) | Methods and apparatus for noise reduction in digital images | |
Singh et al. | A novel dehazing model for remote sensing images | |
US10198801B2 (en) | Image enhancement using self-examples and external examples | |
JP6390173B2 (ja) | 情報処理装置、情報処理システム、画像処理方法、及び、プログラム | |
US7751641B2 (en) | Method and system for digital image enhancement | |
US11145032B2 (en) | Image processing apparatus, method and storage medium for reducing color noise and false color | |
Li et al. | Single image super-resolution using combined total variation regularization by split Bregman Iteration | |
DE102020133244A1 (de) | Kantenbewusste Aufwärtsskalierung für verbesserte Bildschirminhaltsqualität | |
Choi et al. | Resolution improvement of infrared images using visible image information | |
Muhammad et al. | Frequency component vectorisation for image dehazing | |
Teodoro et al. | Sharpening hyperspectral images using plug-and-play priors | |
US20150302273A1 (en) | Image processing device, image processing method and medium | |
Deng et al. | Efficient test-time adaptation for super-resolution with second-order degradation and reconstruction | |
WO2016051716A1 (ja) | 画像処理方法、画像処理装置、及び画像処理プログラムを記憶する記録媒体 | |
Mandal et al. | Employing structural and statistical information to learn dictionary (s) for single image super-resolution in sparse domain | |
Chen et al. | Face super resolution based on parent patch prior for VLQ scenarios | |
US8761507B1 (en) | Bilateral filter optimization | |
Pang et al. | Image restoration via the adaptive TVp regularization | |
Pal et al. | Visibility enhancement of fog degraded images using adaptive defogging function | |
Miyata | Inter-channel relation based vectorial total variation for color image recovery | |
Chang et al. | Super resolution using trilateral filter regression interpolation | |
Su et al. | Single image super-resolution based on space structure learning |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
121 | Ep: the epo has been informed by wipo that ep was designated in this application |
Ref document number: 13857754 Country of ref document: EP Kind code of ref document: A1 |
|
ENP | Entry into the national phase |
Ref document number: 2014549833 Country of ref document: JP Kind code of ref document: A |
|
WWE | Wipo information: entry into national phase |
Ref document number: 2013857754 Country of ref document: EP |
|
WWE | Wipo information: entry into national phase |
Ref document number: 14647531 Country of ref document: US |
|
NENP | Non-entry into the national phase |
Ref country code: DE |