A remote sensing image light and color uniformity method for optimizing brightness, contrast and color difference

Abstract

The application discloses a remote sensing image light and color uniformization method for optimizing brightness, contrast and color difference, and belongs to the color correction technical field in image processing. The scheme comprises the following steps: firstly, converting the collected remote sensing image from an RGB color space to a YCbCr color space; then, correcting the brightness and the uneven contrast distribution phenomenon of the Y channel by adopting a block Wallis transformation method; for the Cb and Cr channels, taking a parameterized spline curve as a color remapping function, and using the calculated color corresponding relationship in the image overlapping area to constrain the color remapping function, solving the parameters of the curve by adopting a convex quadratic programming, and correcting the color difference between images by using the color remapping function; and finally, converting the remote sensing image from the YCbCr color space to the RGB color space. The application respectively performs brightness and color correction in different channels of the image, can improve the uneven brightness and contrast distribution in a single image, and can effectively reduce the color difference between multiple images.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of color correction in image processing, in particular to a remote sensing image dodging and color uniformity method for optimizing brightness, contrast and color difference. BACKGROUND

[0002] Optical satellites are easily affected by factors such as seasonal and atmospheric changes, illumination and sensor distortion. Therefore, satellite images covering a large area often have obvious color differences. In recent years, with the rapid development of earth observation, data dodging and splicing using multi-view remote sensing images can produce a more comprehensive, reliable and complete panoramic image for earth observation. The existence of image color difference not only affects the visual effect of the image, but also increases the difficulty of image splicing and reduces the quality of the spliced image. In the process of obtaining a panoramic image, it is particularly important to eliminate the color difference between images and improve the quality of the spliced image.

[0003] Existing dodging and color uniformity algorithms have their own advantages, but they are specific and need to be adjusted manually in practical applications to achieve satisfactory results. In addition, dodging and color uniformity processing can easily cause information loss and degrade image quality. Due to the complexity of the remote sensing image formation process, existing dodging and color uniformity models often use flexible spline curves as color remapping functions.

[0004] A spline curve is a piecewise interpolation function, and each local curve segment is determined by only three consecutive anchor points. The commonly used color uniformity algorithm uses a parameterized spline curve as a remapping function, represents the color difference between multiple images as an energy function, then optimizes and solves the energy function, and finally performs color correction according to the remapping function. This type of algorithm can effectively eliminate the color difference between multiple images, but it cannot eliminate the inconsistency of brightness and contrast within the image at the same time. SUMMARY

[0005] To solve the above problems, the present application provides a new type of remote sensing image dodging and color uniformity method, which solves the problem that existing dodging and color uniformity algorithms cannot simultaneously correct the brightness and contrast of the image when correcting the color difference of the image, and improves the quality of the resulting image

[0006] To achieve the above purpose, the present application provides a remote sensing image dodging and color uniformity method for optimizing brightness, contrast and color difference, comprising the following steps:

[0007] (1) converting the collected remote sensing image from the RGB color space to the YCbCr color space;

[0008] (2) using the block Wallis transformation method to correct the brightness and contrast distribution of the Y channel of the remote sensing image;

[0009] (3) For the Cb, Cr channels of the remote sensing images, a parameterized spline curve is used as a color remapping function, and the color correspondence calculated in the overlapping area of the images is used to constrain the curve, and the parameters of the curve are solved by using convex quadratic programming, and the color remapping function is used to correct the color difference between the images.

[0010] (4) The remote sensing images are converted from the YCbCr color space to the RGB color space to restore the images.

[0011] Further, the block Wallis transformation method in step 2 comprises:

[0012] (2.1) Determine the image processing order: construct an adjacency graph according to the coordinate information of the remote sensing images, treat the remote sensing images as nodes of the adjacency graph, perform a breadth-first search on the adjacency graph, and take the order of visiting the nodes in the graph as the order of processing the images;

[0013] (2.2) Determine the current reference image of the remote sensing image: according to the order of visiting the nodes in the breadth-first search, take the predecessor node of the node as the current reference image of the remote sensing image;

[0014] (2.3) Determine the original reference image of the remote sensing image: calculate the average gradient of the Y channel of all remote sensing images to be processed, and take the image with the largest average gradient as the original reference image of all remote sensing images;

[0015]

[0016] Δ x = I(x+1, y) - I(x, y)

[0017] Δ y = I(x, y+1) - I(x, y)

[0018] Wherein: G is the average gradient, W and H are the width and height of the remote sensing image respectively; I(x, y) is the gray value of the pixel point (x, y) of the remote sensing image;

[0019] (2.4) Determine the reference mean and standard deviation of the remote sensing image: weight the mean and standard deviation of the current reference image and the original reference image by 1:1, and obtain the reference mean and reference standard deviation of the Y channel of the remote sensing image;

[0020] (2.5) Determine the number of blocks of the remote sensing image: according to the preset width and height w and h of the image block, divide the image block, and calculate the number of blocks in the row and column directions of the Y channel of the remote sensing image:

[0021]

[0022] Wherein: CV is the coefficient of variation of the remote sensing image; CV ref is the coefficient of variation of the current reference image; r and c are the number of rows and columns of the remote sensing image respectively, and W and H are the number of blocks in the row and column directions of the Y channel of the remote sensing image respectively;

[0023] Then, for the divided image blocks, the mean and standard deviation are calculated;

[0024] (2.6) Determine the image block corner point parameters: according to the number of image blocks to which the corner point belongs, the mean of the mean of the image blocks and the mean of the standard deviation are assigned to the corner point, and when the mean or standard deviation of the image block is zero, the image block does not participate in the calculation of the corner point parameters;

[0025] (2.7) Determine the mean and standard deviation of the image elements in the image block: according to the mean and standard deviation of the image block corner points, the mean and standard deviation of the image elements in the image block are calculated by using the bilinear interpolation method;

[0026] (2.8) Wallis transformation: for each image element of the image block, Wallis transformation is performed to complete the uniform light processing of the Y channel of the remote sensing image;

[0027]

[0028] Wherein, g(x, y) is the gray value of the original image pixel point (x, y); m g and s g are the mean and standard deviation of the original image respectively; f(x, y) is the gray value of the result image pixel point (x, y); m ref and s ref are the reference mean and reference standard deviation of the Y channel of the remote sensing image respectively.

[0029] Further, in step (2.5) of the block Wallis transformation method, for the remote sensing image with a width and height of about 8000 pixels, the w and h of the image block are both set to 512, and for other sizes of the remote sensing image, the size of the image block is scaled in proportion according to the ratio of its own width and height to 8000.

[0030] Further, in step (2.6) of the block Wallis transformation method, the number of image blocks to which the corner point belongs: for the four corner points of the image block, there are:

[0031] The first case is that the corner point belongs to only one image block;

[0032] The second case is that the corner point is shared by two image blocks;

[0033] The third case is that the corner point is shared by four image blocks.

[0034] Further, the parameterized spline curve comprises:

[0035] According to the color gamut range of the Cb and Cr channels of the remote sensing image, N equidistant anchor points are set And Then, two parameterized spline curves are constructed as color remapping curves according to the anchor points, so that they can cover the intensity range of the Cb and Cr channels of the remote sensing image respectively, And are known parameters, And are unknown parameters, and the parameterized spline curves are parameterized by unknown parameters And .

[0036] Further, the constraints of the color correspondence relationship in step 3 comprise:

[0037] The cumulative distribution functions CDF of the Cb and Cr channels in the overlapping area are calculated respectively, and the equal-probability quantiles corresponding to the two CDFs are taken as the color correspondence relationship The remapping values are expressed by unknown parameters And . That is,

[0038]

[0039]

[0040] Where t∈[0,1] is an interpolation coefficient.

[0041] Further, the step 3 of solving the parameters of the curve by using convex quadratic programming comprises:

[0042] (3.1) Calculate And the remapped color values And to construct an energy function E;

[0043]

[0044] Where: w ij is a weight, which is proportional to the area of the overlapping area; K is the number of anchor points of the remapping curve, K=N; M is the number of color correspondence relationships extracted in the image overlapping area,

[0045] and the parameterized spline curve f i and f j is constrained by the upper and lower bounds of the slope of the parameterized spline curve and the value range of the color remapping:

[0046]

[0047]

[0048] wherein: τ l and τ u are the lower and upper bounds of the slope of the parameterized spline curve;[v min ,v max are the color gamut ranges of the color remapping in the Cb and Cr channels;

[0049] (3.2) minimizing the energy function with convex quadratic programming to solve the parameters of the parameterized spline curve and and correcting the color difference of the Cb and Cr channels of the image through the parameterized spline curve.

[0050] The beneficial effects of the present application are:

[0051] 1. The use of the block Wallis transform algorithm to correct the Y channel of the image can effectively reduce the uneven distribution of light and dark within a single image. At the same time, the algorithm can adjust the inconsistent contrast of the image to balance the contrast within the image and improve the quality of the image.

[0052] 2. The flexibility of the spline curve model enables it to effectively represent the color difference between remote sensing images. Therefore, using the model to even the color of the Cb and Cr channels of the image can effectively correct large color differences.

[0053] 3. The brightness correction and color correction in different channels of the image can improve the uneven distribution of brightness and contrast within a single image and effectively reduce the color difference between multiple images. BRIEF DESCRIPTION OF DRAWINGS

[0054] Figure 1 is the overall structure diagram of the remote sensing image light and color evenness method for optimizing brightness, contrast and color difference according to the embodiment of the present application.

[0055] Figure 2 is the schematic diagram of the color remapping curve of the segmented quadratic spline curve modeling according to the embodiment of the present application.

[0056] Figure 3 is the schematic diagram of the two CDF curves in the overlapping area of the image according to the embodiment of the present application.

[0057] Figure 4 is a schematic diagram of experimental results of embodiments of the present application on Landsat-8 data sets. DETAILED DESCRIPTION

[0058] The specific embodiments of the present application will be described in detail below with reference to the accompanying drawings and preferred embodiments.

[0059] As shown in the drawings, Figure 1 the embodiments of the present application provide a remote sensing image dodging and color uniformization method for optimizing brightness, contrast and color difference, comprising the following steps:

[0060] S101, converting the collected remote sensing image from RGB color space to YCbCr color space;

[0061] S102, using the block Wallis transformation method to correct the brightness and contrast distribution of the Y channel of the remote sensing image.

[0062] comprising the following steps:

[0063] (1) First, according to the geographic coordinate information of the input image, an adjacency graph is constructed to record the overlap relationship between the images, and then a breadth-first search is performed on the adjacency graph, taking the order of visiting nodes as the order of image dodging processing, and taking the predecessor node of the node as the current reference image of the image.

[0064] (2) Calculate the average gradient of the Y channel of all images to be processed by the formula, and take the image with the maximum average gradient value as the original reference image of all images.

[0065]

[0066] Δ x = I(x+1, y) - I(x, y)

[0067] Δ y = I(x, y+1) - I(x, y)

[0068] Where: G is the average gradient, W and H are the width and height of the remote sensing image respectively; I(x, y) is the gray value of the remote sensing image pixel point (x, y);

[0069] (3) Weighted average the mean and standard deviation of the current reference image and the original reference image according to the weight of 1:1, to obtain the reference mean and reference standard deviation of the Y channel of each remote sensing image.

[0070] (4) According to the image size of the experiment, for the remote sensing image with width and height of about 8000 pixels, the w and h of the image block are set to 512, and for other sizes of remote sensing image, the size of the image block is scaled in proportion according to the ratio of its own width and height to 8000; then the block number of the image Y channel in the row and column directions is calculated by the formula, and the mean and standard deviation of each image block are calculated, and the block number calculation formula is as follows:

[0071]

[0072] Wherein: CV is the coefficient of variation of the remote sensing image; CV ref is the coefficient of variation of the current reference image; r and c are the number of rows and columns of the remote sensing image respectively, and W and H are the block number of the remote sensing image Y channel in the row and column directions.

[0073] (5) According to the number of image blocks to which the corner point belongs:

[0074] The first case is that the corner point belongs to only one image block.

[0075] The second case is that the corner point is shared by two image blocks.

[0076] The third case is that the corner point is shared by four image blocks.

[0077] The mean of the mean of the image blocks and the mean of the standard deviation are assigned to the corner point, and when the parameter (mean or standard deviation) of the image block is zero, the image block does not participate in the calculation of the corner point parameter.

[0078] (6) According to the mean and standard deviation of the image block corner point, the bilinear interpolation method is used to calculate the mean and standard deviation of the image block.

[0079] (7) For each image block pixel, Wallis transformation is performed by using the formula to complete the uniform light processing of the image Y channel.

[0080]

[0081] Wherein, g(x, y) is the gray value of the original image pixel point (x, y); m g and s g are the mean and standard deviation of the original image respectively; f(x, y) is the gray value of the result image pixel point (x, y); m ref and s ref are the reference mean and reference standard deviation of the remote sensing image Y channel respectively.

[0082] S103. For the Cb and Cr channels of the remote sensing image, the parameterized spline curve is used as the color remapping function, and the color correspondence calculated in the overlapping area of ​​the image is used as a constraint. The parameters of the curve are solved by convex quadratic programming, and the color remapping function is used to correct the color difference between the images.

[0083] (1) Parametric spline curves:

[0084] like Figure 2 As shown, based on the color gamut range of the Cb and Cr channels in the remote sensing image, N equally spaced anchor points (intersection points in the figure) are set respectively. and In the experiment, N=6; then, two parameterized spline curves were constructed based on the anchor points as color remapping curves, so that they could respectively cover the intensity range of the Cb and Cr channels of the remote sensing image. and Given parameters, and For unknown parameters, the spline curves are respectively composed of unknown parameters. and Perform parameterized representation.

[0085] (2) Constraints on color correspondence

[0086] like Figure 3 As shown, the cumulative distribution functions (CDF) of the Cb and Cr channels within the overlapping area of ​​the images are calculated respectively, and then the equal probability quantiles corresponding to the two CDF curves are taken as the color correspondence. Using unknown parameters respectively and To express remapping values Right now

[0087]

[0088]

[0089] Where t∈[0,1] are the interpolation coefficients.

[0090] In the experiment, the probability values ​​were as follows: Thus, 16 color correspondences can be obtained in each overlapping area.

[0091] (3) Solve the parameters of the curve using convex quadratic programming.

[0092] Calculated from the parameterized spline curve and Remapped color values and To construct the energy function E;

[0093]

[0094] wherein: w ij is a weight, the value of which is proportional to the area of the overlapping region; K is the number of anchor points of the remapping curve, K=N, M is the number of color correspondence extracted in the image overlapping region, In the experiment, K=N=6, M=16.

[0095] and the upper and lower limits of the slope of the parameterized spline curve f i and f j and the value range of color remapping are constrained:

[0096]

[0097]

[0098] wherein: τ l and τ u are the lower and upper limits of the slope of the parameterized spline curve;[v min ,v max ] is the color gamut range of color remapping in the Cb and Cr channels.

[0099] In the experiment, the lower limit of the curve slope is set to 0.3, the upper limit of the curve slope is set to 5, and the value range of the result image Cb or Cr channel after color remapping remains unchanged.

[0100] The energy function is minimized by convex quadratic programming to solve the parameters of the parameterized spline curve and and correct the color difference of the image Cb and Cr channels through the parameterized spline curve.

[0101] S104, convert the remote sensing image from YCbCr color space to RGB color space to restore the image.

[0102] The results of the experiment are shown in FIG. 1. Figure 4 To represent the color difference between images, the white box in the figure highlights the area of inconsistent color.

[0103] The evaluation indicators used in performance evaluation are: SSIM (Structural Similarity), QCCED (Quality Considered Color Euclidean Distance), and OEI (One-dimensional Entropy of Images).

[0104] The index SSIM measures the similarity between images from three aspects of brightness, contrast and structure, and the value range is [0, 1], and the greater the value is, the smaller the image distortion is, and the formula is as follows:

[0105] SSIM (X, Y) = l (X, Y) · c (X, Y) · s (X, Y)

[0106]

[0107] Wherein, μ X , μ Y and σ X , σ Y respectively represent the mean and standard deviation of image X and image Y. σ XY represents the covariance of image X and image Y. C1, C2 and C3 are constants, and in order to avoid the denominator being zero, C1 = (k1L) 2 , C2 = (k2L) 2 , Generally, k1 = 0.01, k2 = 0.03, and L = 255.

[0108] The index QCCED includes color Euclidean distance (CED) and average gradient (MG). CED reflects the difference between pixel values of images. The smaller the CED is, the smaller the color difference between images is. MG reflects the clarity of image texture. The higher the MG is, the higher the quality of the resulting image is. Then the index QCCED can be calculated using the following formula, wherein c is a constant that remains stable; generally it is set to 1.

[0109]

[0110]

[0111]

[0112] The index OEI is a statistical form of a feature, which reflects how much the average information in the image is, and indicates the information amount contained by the aggregation feature of the image gray scale distribution. The formula is as follows:

[0113]

[0114] Wherein, p i represents the proportion of pixels with the gray scale value i in the image. The greater the one-dimensional entropy of the image is, the greater the information amount of the image is, and at the same time, the richer the gray scale level contained by the image is.

[0115] By comparing with the existing two uniform light and color algorithms, it is found that the method proposed in the application has the best performance, and the specific contents are shown in the following table.

[0116]

[0117] Although the specific embodiments of the present application have been described above, it will be understood by those skilled in the art that these are merely examples, and various changes or modifications can be made to these embodiments without departing from the principles and spirit of the present application, and therefore, the scope of protection of the present application is defined by the appended claims.

Claims

1. A method for optimizing brightness, contrast and color difference of remote sensing images, characterized in that, The method comprises the following steps: (1) converting the collected remote sensing image from an RGB color space to a YCbCr color space; (2) correcting the brightness and contrast uneven distribution of the remote sensing image by using a block Wallis transformation method on the Y channel of the remote sensing image; (3) taking a parameterized spline curve as a color remapping function, and using the calculated color correspondence in the image overlap area to constrain the color remapping function, solving the parameters of the curve by using a convex quadratic programming, and correcting the color difference between the images by using the color remapping function; The parameterized spline curve comprises: setting N equidistant anchor points according to the color gamut range of the Cb and Cr channels of the remote sensing image and Then, two parameterized spline curves are constructed as the color remapping curves according to the anchor points, so that the two parameterized spline curves can cover the intensity range of the Cb and Cr channels of the remote sensing image, respectively and are known parameters, and are unknown parameters, and the parameterized spline curve is parameterized by the unknown parameters and respectively. The constraint of color correspondence includes: calculating cumulative distribution function (CDF) of Cb and Cr channels in the overlapping area respectively, taking equal probability quantile corresponding to the two CDFs as color correspondence; , }; unknown parameters and are used to express remapping value , , i.e. , ; wherein is an interpolation coefficient; Solving the parameters of the curve by using a convex quadratic programming comprises: (3.1) calculating, by means of the parameterized spline curve and the remapped color values and to construct an energy function E; wherein: is a weight, the value being proportional to the area of the overlapping region; , K is the number of anchor points of the remapping curve K=N, M is the number of color correspondences extracted in the image overlapping region, ; and the upper and lower limits of the slope of the parametric spline curve and and the value range of the color remapping are constrained: wherein: and are lower and upper limits of the parametric spline curve slope; , are color gamut ranges of color remapping in the Cb and Cr channels; (3.2) minimizing the energy function with convex quadratic programming to solve the parameters of the parameterized spline curve and and correcting color difference of image Cb and Cr channels through the parameterized spline curve; (4) converting the remote sensing image from the YCbCr color space to the RGB color space to restore the image.

2. The method for optimizing brightness, contrast and color difference of remote sensing image according to claim 1, characterized in that, The block Wallis transformation method in step (2) comprises: (2.1) determining an image processing order: constructing an adjacency graph according to the coordinate information of the remote sensing image, taking the remote sensing image as a node of the adjacency graph, performing a breadth-first search on the adjacency graph, and taking the order of visiting the nodes in the graph as the order of processing the images; (2.2) determining a current reference image of the remote sensing image: taking the predecessor node of the node as the current reference image of the remote sensing image according to the order of visiting the nodes in the breadth-first search; (2.3) determining an original reference image of the remote sensing image: calculating the average gradient of the Y channel of all the remote sensing images to be processed, and taking the image with the largest average gradient as the original reference image of all the remote sensing images; wherein: G is the average gradient of the remote sensing image, W and H are the width and height of the remote sensing image respectively; is the gray value of the pixel point (x, y) of the remote sensing image; (2.4) determining a reference mean value and a reference standard deviation of the remote sensing image: weighting the mean value and the standard deviation of the current reference image and the original reference image according to a weight ratio of 1:1 to obtain the reference mean value and the reference standard deviation of the Y channel of the remote sensing image; (2.5) determining the number of blocks of the remote sensing image: according to a preset width and height of the image block w and h the number of blocks in the row and column directions of the Y channel of the remote sensing image is calculated by the following formula: wherein: is a coefficient of variation of the remote sensing image; is a coefficient of variation of the current reference image; r and c are the number of rows and columns of the remote sensing image, respectively, and W and H are the number of image blocks in the row and column directions of the Y channel of the remote sensing image, respectively. Then, the mean value and the standard deviation of each image block are calculated. (2.6) determining the corner point parameters of the image block: according to the number of image blocks to which the corner point belongs, the mean value of the mean value of the image blocks and the mean value of the standard deviation of the image blocks are assigned to the corner point, and when the mean value or the standard deviation of the image block is zero, the image block does not participate in the calculation of the corner point parameters; (2.7) determining the mean value and the standard deviation of the image elements in the image block: the mean value and the standard deviation of the image elements in the image block are calculated by using a bilinear interpolation method according to the mean value and the standard deviation of the corner points of the image block; (2.8) performing Wallis transformation: for each image element of the image block, Wallis transformation is performed to complete the uniform light processing of the Y channel of the remote sensing image; wherein, is the gray value of the original image pixel point (x, y); and are the mean value and the standard deviation of the original image respectively; is the gray value of the result image pixel point (x, y); and are the reference mean value and the reference standard deviation of the Y channel of the remote sensing image respectively.

3. The method for optimizing brightness, contrast and color difference of remote sensing image dodging and color matching according to claim 2, characterized in that, In step (2.5) of the block Wallis transformation method, for the remote sensing image with a width and a height of about 8000 pixels, the w and h of the image block are both set to 512, and for other sizes of the remote sensing image, the size of the image block is scaled in proportion according to the ratio of its own width and height to 8000.

4. The method for optimizing brightness, contrast and color difference of remote sensing image according to claim 2, characterized in that, In step (2.6) of the block Wallis transformation method, the number of image blocks to which the corner point belongs: for the four corner points of the image block, there are two cases: The first case is that the corner point only belongs to one image block; The second case is that the corner point is shared by two image blocks; The third case is that the corner point is shared by four image blocks.

Images