Adaptive weight allocation chromosome stripe contrast enhancement method and system

Abstract

The invention provides an adaptive weight distribution chromosome stripe contrast enhancement method and system, and the method comprises the steps: firstly improving the significance of a chromosome stripe structure and an edge contour in a chromosome gray image through an edge enhancement convolution operator; carrying out chromosome monomer target detection and instance segmentation on the metaphase chromosome based on an instance segmentation network of deep learning to obtain a chromosome monomer ROI region; then, extracting a structure tensor response of the ROI region of the chromosome monomer, designing an adaptive enhancement weight of a histogram merging interval based on a normalized difference between a response value and global statistical information, and performing reconstruction and interpolation on a histogram of the image of the ROI region of the chromosome monomer; and finally, adaptively enhancing the contrast of the chromosome stripe through iterative histogram matching. According to the method, the original image texture structure is kept, and meanwhile, the definition and the distinguishability of chromosome stripes are effectively and adaptively improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of chromosome image processing, and in particular, to a chromosome band contrast enhancement method and system with adaptive weight distribution. BACKGROUND

[0002] Chromosomes are the carriers of genetic information in the nucleus of a cell, appearing as filamentous or rod-like structures under a microscope. They play a crucial role in the process of cell division, carrying and transmitting genetic information. The morphology and structure of chromosomes are of great significance in genetic research, disease diagnosis, and cell biology studies. Chromosome images are usually obtained through microscopy, but these images often have low contrast and unclear details, which poses difficulties for chromosome analysis and diagnosis. Therefore, improving the contrast of chromosome images and enhancing their detailed information are crucial for accurately analyzing chromosome structure and function. Chromosome contrast enhancement can make the morphology and structure of chromosomes more clear, facilitating observation and analysis by scientists and medical experts. This has important value for tasks such as chromosome counting, morphological classification, and structural anomaly detection, helping to improve the accuracy and efficiency of diagnosis and having important significance for the diagnosis and research of genetic diseases.

[0003] Existing chromosome contrast enhancement schemes can meet the requirements of experiments, but still have many shortcomings. For example, patent application document CN115601374A discloses a chromosome image segmentation method, which enhances the gray level and contrast of the image using a histogram normalization image processing method, making the bands inside the chromosome clear and distinguishable. However, this patent does not solve the problem that the band structure inside the chromosome may be ignored in global adjustment, leading to loss of details or insufficient contrast. SUMMARY

[0004] In view of the defects in the prior art, the purpose of the present application is to provide a chromosome band contrast enhancement method and system with adaptive weight distribution.

[0005] According to the chromosome band contrast enhancement method with adaptive weight distribution provided by the present application, the following steps are included:

[0006] Step 1: Apply an edge enhancement convolution operator to the original chromosome gray image after normalization for sharpening preprocessing;

[0007] Step 2: Use an instance segmentation model in deep learning to detect and segment the single body of the chromosome, to obtain the single body ROI region of each chromosome, including the bounding box and the mask;

[0008] Step 3: Linear stretching is performed on the gray value of the chromosome monomer ROI region, and the gray value is mapped to 0 to 255; the horizontal and vertical gradients of the chromosome monomer ROI region are calculated, the structure tensor is constructed, and the structure response value reflecting the texture structure intensity and direction stability of the pixel neighborhood is obtained;

[0009] Step 4: The gray value of the chromosome monomer ROI region is divided into multiple histogram merging intervals, and the structure tensor response mean value of each interval is extracted; based on the deviation and standard deviation of the mean value and the global response, the weight factor of each merging interval is calculated; based on the weight factor, the value of the original histogram merging interval is adjusted, and finally the complete target histogram is obtained by interpolation;

[0010] Step 5: The image of the chromosome monomer ROI region is enhanced by iterative histogram matching, and the enhancement effect is evaluated by using the contrast improvement ratio CIR index, and the iteration is stopped when the difference of CIR is less than the set threshold for two consecutive times, and it is determined that the contrast enhancement of the chromosome monomer has reached the expected requirement.

[0011] Preferably, the step 1 comprises: using a center enhancement and periphery neutralization edge enhancement convolution kernel to enhance the edges and details of the image while suppressing smooth areas; the edge enhancement convolution kernel sharpens the intermediate chromosome gray image, highlighting the strip structure and edge information of the chromosome monomer.

[0012] Preferably, the step 2 comprises: inputting the original chromosome gray image into the instance segmentation model Mask-RCNN, which identifies, locates and distinguishes each individual chromosome by learning the local features in the image, and generates the corresponding mask, bounding box and confidence score; the mask is used to accurately mark the chromosome region to provide the ROI region for subsequent chromosome contrast; the bounding box is used to represent the position of each chromosome; the confidence score is used to evaluate the confidence degree of the model to the detection result, and the higher the score, the greater the probability that the instance is a chromosome.

[0013] Preferably, the step 3 comprises:

[0014] linear stretching is performed on the gray value of the chromosome monomer ROI region, and the gray value is mapped to 0 to 255; the horizontal and vertical gradients of the chromosome monomer ROI region are calculated, the structure tensor is constructed, and the structure response value reflecting the texture structure intensity and direction stability of the pixel neighborhood is obtained; x 、I y , three basic components of the structure tensor are constructed by combining the gradient map pixel by pixel, that is, the horizontal gradient square term I xx , the vertical gradient square term I yy and the product term I xy of horizontal and vertical gradients, and the expression is:

[0015] I xx =I x2

[0016] I yy = I y 2

[0017] I xy = I x × I y

[0018] The Gaussian filter is used to smooth each component of the tensor to obtain the structure tensor M, and the structure response value R representing the stability of the intensity and direction of the texture structure in the neighborhood of each pixel is obtained, and the expression is:

[0019]

[0020] R = det(M) - k × (trace(M)) 2

[0021] wherein det(M) is the determinant of M, trace(M) is the trace of M, and k is an adjustment factor;

[0022] After normalization, the structure response map of the chromosome monomer is obtained by taking R as a local adjustment factor.

[0023] Preferably, the step 4 comprises:

[0024] calculating the global mean R of the structure response map mean and the standard deviation R std dividing the gray scale range of the chromosome image into J merged intervals, and extracting the structure tensor response mean R j of the corresponding region for each merged interval, 1≤j≤J; according to the deviation of the response value from the global mean R mean of the structure response map, the standard deviation is normalized to calculate the weight factor weight of the interval, and the weight factor is limited within 0.5-2.0, and the expression is:

[0025]

[0026] The weight factor is used to adjust the merged interval value of the original histogram, and if the current merged interval is not the last merged interval, i.e. j

[0027]

[0028] wherein N is a normalization parameter for adjusting the change amount of each step;

[0029] Then, calculate the histogram values of the two parts before and after each histogram merging interval;

[0030] When H[j] < H[j + 1], keep the histogram values of the center points C j , C j+1 unchanged, increase the histogram value of the j right - hand side point of C j+1 and decrease the histogram value of the left - hand side point

[0031] of C j , and add each point to the interpolation list l; j+1 l(C

[0032]

[0033] j When H[j] > H[j + 1], keep the histogram values of the center points C, C j+1 unchanged, decrease the histogram value of the j ] right - hand side point of C j+1 and increase the histogram value of the left - hand side point

[0034] of C j , and add each point to the interpolation list l; j+1 l(C

[0035] <00002]7>

[0036] If the current merging interval is the last interval, that is, j = J, then directly multiply the interval center C j by the weight factor weight and add it to the interpolation list l;

[0037] Use the linear interpolation method to generate a complete 256 - level gray - scale histogram, and normalize the interpolated histogram to ensure that the total number of pixels is the same as the total number of pixels in the ROI area of the chromosome monomer image.

[0038] According to the adaptive weight - assignment chromosome band contrast enhancement system provided by the present invention, it includes:

[0039] Module M1: Apply the edge - enhancement convolution operator to the original chromosome gray - scale image after normalization for sharp - ening pre - processing;

[0040] Module M2: using instance segmentation model in deep learning to detect and segment the chromosome monomer, to obtain the monomer ROI region of each chromosome, including the bounding box and the mask;

[0041] Module M3: linear stretching the gray value of the chromosome monomer ROI region, mapping the gray value to 0 to 255; calculating the horizontal and vertical gradients of the chromosome monomer ROI region, constructing the structure tensor, and obtaining the structure response value reflecting the texture structure intensity and direction stability of the pixel neighborhood;

[0042] Module M4: dividing the gray value of the chromosome monomer ROI region into multiple histogram merging intervals, extracting the structure tensor response mean value of each interval; based on the deviation and standard deviation of the mean value and the global response, calculating the weight factor of each merging interval; based on the weight factor, adjusting the value of the original histogram merging interval, and finally interpolating to obtain the complete target histogram;

[0043] Module M5: image enhancement of the chromosome monomer ROI region by iterative histogram matching, using the contrast improvement ratio CIR index to evaluate the enhancement effect, and stopping iteration when the difference of CIR for two consecutive times is lower than the set threshold, determining that the contrast enhancement of the chromosome monomer has reached the expected requirement.

[0044] Preferably, the module M1 includes: using a center enhancement and periphery neutralization edge enhancement convolution kernel to enhance the edges and details of the image while suppressing smooth areas, the edge enhancement convolution kernel performs sharpening operation on the intermediate chromosome gray image, highlighting the strip structure and edge information of the chromosome monomer.

[0045] Preferably, the module M2 includes: inputting the original chromosome gray image into the instance segmentation model Mask-RCNN, which identifies, locates and distinguishes each individual chromosome by learning the local features in the image, and generates the corresponding mask, bounding box and confidence score; the mask is used to accurately mark the chromosome region to provide the ROI region for subsequent chromosome contrast; the bounding box is used to represent the position of each chromosome; the confidence score is used to evaluate the confidence degree of the model to the detection result, and the higher the score, the greater the probability that the instance is a chromosome.

[0046] Preferably, the module M3 includes:

[0047] linear stretching the gray value of the chromosome monomer ROI region, mapping the gray value to 0 to 255; calculating the horizontal and vertical gradients of the chromosome monomer ROI region, constructing the structure tensor, and obtaining the structure response value reflecting the texture structure intensity and direction stability of the pixel neighborhood; x y , constructing three basic components of the structure tensor by combining the gradient map pixel by pixel, namely the horizontal gradient square term I xx , the vertical gradient square term I​yy and the product term I of horizontal and vertical gradients xy , the expression is:

[0048] I xx = I x 2

[0049] I yy = I y 2

[0050] I xy = I x × I y

[0051] The tensor components are smoothed by using Gaussian filter to obtain the structure tensor M, and the structure response value R representing the stability of the intensity and direction of the texture structure in the neighborhood of each pixel is obtained, the expression is:

[0052]

[0053] R = det(M) - k × (trace(M)) 2

[0054] Wherein, det(M) is the determinant of M, trace(M) is the trace of M, and k is the adjustment factor;

[0055] After normalization, the structure response map of the chromosome monomer is obtained by taking R as the local adjustment factor.

[0056] Preferably, the module M4 comprises:

[0057] Calculate the global mean R of the structure response map mean and the standard deviation R std , the gray scale range of the chromosome image is divided into J merging intervals, and the structure tensor response mean R j , 1≤j≤J; according to the deviation of the response value from the global mean R mean of the structure response map, the standard deviation is normalized to calculate the weight factor weight of the interval, and the weight factor is limited within 0.5-2.0, the expression is:

[0058]

[0059] The weight factor is used to adjust the merging interval value of the original histogram, if the current merging interval is not the last merging interval, i.e. j

[0060]

[0061] wherein N is a normalization parameter, used to adjust the amount of change at each step;

[0062] Then, the histogram values of the two parts before and after each histogram merging interval are calculated.

[0063] When H[j]<H[j+1], the histogram values of the center points C j , C j+1 of each merging interval are kept unchanged, the histogram value of the first point to the right of C j is increased and the histogram value of the first point to the left of C j+1 is decreased, and each point is added to the interpolation list l.

[0064] l(C j )=H[j], l(C j+1 )=H[j+1]

[0065]

[0066] When H[j]<H[j+1], the histogram values of the center points C j , C j+1 of each merging interval are kept unchanged, the histogram value of the first point to the right of C j is increased and the histogram value of the first point to the left of C j+1 is decreased, and each point is added to the interpolation list l.

[0067] l(C j )=H[j], l(C j+1 )=H[j+1]

[0068]

[0069] If the current merging interval is the last interval, i.e. j=J, then the interval center C j is directly added to the interpolation list l after being multiplied by the weight factor weight.

[0070] Using the linear interpolation method, a complete 256-level grayscale histogram is generated, and the interpolated histogram is normalized to ensure that the total pixel number is consistent with the total pixel number of the ROI region of the chromosome monomer image.

[0071] Compared with the prior art, the present application has the following beneficial effects:

[0072] This invention provides a method for enhancing chromosome band contrast based on adaptive weight allocation using structure tensor awareness. This method can better preserve detailed information, avoid image distortion caused by over-enhancement, and achieve precise contrast enhancement of chromosome images, effectively improving the visual effect and analytical usability of the images. This invention not only improves the contrast and clarity of chromosome images, but also enhances the algorithm's adaptability to different types of chromosome images, providing more reliable technical support for genetic research and clinical diagnosis. Attached Figure Description

[0073] Other features, objects, and advantages of the present invention will become more apparent from the following detailed description of non-limiting embodiments with reference to the accompanying drawings:

[0074] Figure 1 A flowchart of a structure tensor-aware adaptive weight allocation algorithm for enhancing chromosome band contrast.

[0075] Figure 2 A flowchart for adjusting the chromosome monomer histogram for adaptive weight allocation;

[0076] Figure 3a and Figure 3b A schematic diagram of chromosome monomer histogram adjustment for adaptive weight allocation. Detailed Implementation

[0077] The present invention will now be described in detail with reference to specific embodiments. These embodiments will help those skilled in the art to further understand the present invention, but do not limit the invention in any way. It should be noted that those skilled in the art can make several changes and improvements without departing from the concept of the present invention. These all fall within the protection scope of the present invention.

[0078] Example 1

[0079] like Figure 1 As shown in the figure, this embodiment provides an adaptive weight allocation algorithm for enhancing chromosome band contrast based on structure tensor awareness, which includes the following steps:

[0080] S100: Chromosome grayscale image sharpening preprocessing:

[0081] Specifically, after obtaining the grayscale image of the metaphase chromosome karyotype, a custom 5x5 sharpening convolution kernel S is used to perform a convolution operation on the metaphase chromosome karyotype image. The expression is:

[0082]

[0083] Convolution enhances the edge and detail features of the image, providing a clearer image foundation for subsequent structural tensor calculations and contrast enhancement;

[0084] S200: Target detection and instance segmentation of chromosome monomers:

[0085] Specifically, the chromosome karyotype image dataset is input into the instance segmentation model Mask-RCNN, which learns the local features in the image to identify, locate and distinguish each individual chromosome, and generates the corresponding mask, bounding box and confidence score; the mask is used to accurately mark the chromosome region to provide the ROI region for subsequent chromosome comparison; the bounding box is used to represent the position of each chromosome; the confidence score is used to evaluate the confidence of the model on the detection result, and the higher the score, the more likely it is a chromosome;

[0086] S300: Chromosome monomer structure tensor response extraction:

[0087] Specifically, the ROI region of the chromosome monomer image is linearly stretched, and the minimum value p min and the maximum value p max of all pixel values in the ROI are calculated; min The pixel value p is mapped from the original range [p max , p x ] to [0, 255];

[0088]

[0089] The horizontal and vertical gradients I y , I xx are calculated respectively, and the three basic components of the structure tensor, i.e. the horizontal gradient square term I yy , the vertical gradient square term I xy and the product term I xx of the horizontal and vertical gradients, are constructed by pixel-by-pixel combination of the gradient map.

[0090] I x = I 2

[0091] I yy = I y 2

[0092] I xy = I x × I y

[0093] Gaussian filtering is used to smooth each component of the tensor to obtain the structure tensor M, and the structure response value R representing the intensity and direction stability of the texture structure in the neighborhood of each pixel is obtained.

[0094]

[0095] R = det(M) - k x (trace(M)) 2

[0096] where det(M) is the M determinant, trace(M) is the trace of M, k = 0.04;

[0097] After normalization, R is taken as the local adjustment factor to obtain the structural response map of the ROI region of the chromosome monomer image;

[0098] S400: adaptive weight distribution of chromosome monomer histogram adjustment:

[0099] Specifically, the global mean R of the structural response map of the ROI region of the chromosome monomer image is calculated mean and the standard deviation R std The gray scale range of the ROI region of the chromosome monomer image is divided into J = 8 merged intervals with a width L of 32, and the structural tensor response mean R of the corresponding region is extracted for each interval j , 1≤j≤8; According to the deviation of the response value from the global structural response R mean , combined with the standard deviation, the weight factor weight of the merged interval is calculated after normalization processing, and the weight is limited in the range of 0.5 to 2.0 to avoid excessive weight influence;

[0100]

[0101] where ratio and delta are process parameters;

[0102] Using the weight, the values of each merged interval of the original histogram are adjusted. If the current merged interval is not the last merged interval, i.e. j < J, then according to the change trend of the histogram value, the difference between the adjacent merged intervals H[j], H[j+1] is calculated to obtain the corresponding adjustment value diff;

[0103]

[0104] The normalization parameter N = 7 is used to adjust the change amount of each step;

[0105] Then, the histogram values of the two parts before and after each histogram merged interval are calculated;

[0106] The adaptive weight distribution of the chromosome monomer histogram adjustment is as follows Figure 3a and Figure 3b ;

[0107] When H[j] < H[j+1], the histogram values of the center points C j , C j+1 of each merged interval are kept unchanged, and Cj The right side Points Histogram values ​​and reduction of C j+1 Left side Points The histogram values ​​are calculated, and each point is added to the interpolation list l;

[0108] l(C j )=H[j],l(C j+1 )=H[j+1]

[0109]

[0110] When H[j] > H[j+1], the center point C of each merged interval is maintained. j C j+1 The histogram values ​​remain unchanged, and C is reduced. j The right side Points Histogram values ​​and C j+1 Left side Points The histogram values ​​are calculated, and each point is added to the interpolation list l;

[0111] l(C j )=H[j],l(C j+1 )=H[j+1]

[0112]

[0113] If the current merge interval is the last interval (j=8), then directly multiply the merge interval center C8 by the weight and add it to the interpolation list l;

[0114] l(C8)=H[8]×weight

[0115] A complete 256-level grayscale histogram is generated using linear interpolation. The interpolated histogram is then normalized to ensure that its total number of pixels is consistent with the total number of pixels in the ROI region of the chromosome monomorphic image.

[0116] S500: Iterative histogram matching for enhanced contrast in chromosome regions.

[0117] An iterative histogram matching technique was used to progressively enhance the ROI region image of chromosome monosomy. Each iteration adjusted the grayscale distribution based on the updated histogram to optimize the image contrast. During the iteration process, the contrast improvement ratio (CIR) was introduced as a quantitative evaluation index. The CIR change threshold was set to 2. When the CIR difference between two consecutive iterations dropped below this threshold, the iteration process was automatically terminated, indicating that the contrast enhancement of chromosome monosomy had met the expected requirements.

[0118] The application provides a structure tensor perception-based adaptive weight allocation chromosome band contrast enhancement method, which calculates the structure tensor response of a chromosome image, extracts the texture structure features of the image, and adaptively adjusts the histogram based on the features, so that the chromosome image is accurately enhanced. Compared with the prior art, the problem of excessive uniformity in the enhancement effect in the traditional method is effectively avoided, so that the detail information of the chromosome is better preserved. In addition, through adaptive weight allocation, the problem of over-enhancement that may occur in the traditional method is avoided, so that the enhanced image is more natural and closer to the real chromosome morphology. Finally, the application can significantly improve the visual effect of the chromosome image, making the morphology and structure of the chromosome clearer, which helps to improve the accuracy and efficiency of chromosome analysis, and has important significance for the diagnosis and research of genetic diseases.

[0119] Embodiment 2

[0120] The application also provides an adaptive weight allocation chromosome band contrast enhancement system, which can be realized by executing the flow steps of the adaptive weight allocation chromosome band contrast enhancement method, that is, the adaptive weight allocation chromosome band contrast enhancement method can be understood by those skilled in the art as the preferred embodiment of the adaptive weight allocation chromosome band contrast enhancement system.

[0121] The adaptive weight allocation chromosome band contrast enhancement system provided by the application comprises: a module M1, which is used for performing sharpening preprocessing by applying an edge enhancement convolution operator to an original chromosome gray image after normalization; a module M2, which is used for performing monomer target detection and instance segmentation on the chromosome by using an instance segmentation model in deep learning, so as to obtain a monomer ROI region of each chromosome, which comprises a bounding box and a mask; a module M3, which is used for performing linear stretching on the gray value of the chromosome monomer ROI region, mapping the gray value to 0 to 255, calculating the horizontal and vertical gradients of the chromosome monomer ROI region, constructing a structure tensor, and obtaining a structure response value reflecting the texture structure intensity and direction stability of a pixel neighborhood; a module M4, which is used for dividing the gray value of the chromosome monomer ROI region into a plurality of histogram merging intervals, extracting the structure tensor response mean value of each interval, calculating the weight factor of each merging interval based on the deviation and standard deviation of the mean value and the global response, adjusting the value of the original histogram merging interval based on the weight factor, and finally interpolating to obtain a complete target histogram; and a module M5, which is used for performing image enhancement on the chromosome monomer ROI region by iterative histogram matching, evaluating the enhancement effect by using a contrast improvement ratio CIR index, and stopping iteration when the difference between two consecutive CIR values is lower than a set threshold value, so as to determine that the chromosome monomer contrast enhancement has reached the expected requirement.

[0122] The module M1 includes an edge enhancement kernel with central enhancement and peripheral neutralization, which enhances the edges and details of the image while suppressing smooth areas. The edge enhancement kernel performs a sharpening operation on the intermediate chromosome gray image, highlighting the band structure and edge information of the chromosome monomer.

[0123] The module M2 includes inputting the original chromosome gray image into the instance segmentation model Mask-RCNN. The model learns the local features in the image to identify, locate and distinguish each individual chromosome, and generates corresponding masks, bounding boxes and confidence scores. The mask is used to accurately mark the chromosome region to provide the ROI area for subsequent chromosome comparison. The bounding box is used to represent the position of each chromosome. The confidence score is used to evaluate the model's confidence in the detection result. The higher the score, the greater the probability that the instance is a chromosome.

[0124] The module M3 includes:

[0125] Linear stretching is performed on the gray value of the chromosome monomer ROI region to map the gray value to 0 to 255. The horizontal and vertical gradients I x 、I y of the chromosome monomer ROI region are calculated respectively. The three basic components of the structure tensor, namely the horizontal gradient square term I xx , the vertical gradient square term I yy and the product term I xy of the horizontal and vertical gradients, are constructed by pixel-by-pixel combination of the gradient graph, and the expression is:

[0126] I xx = I x 2

[0127] I yy = I y 2

[0128] I xy = I x × I y

[0129] Gaussian filtering is used to smooth each component of the tensor to obtain the structure tensor M, and the structure response value R representing the intensity and direction stability of the texture structure in the neighborhood of each pixel is obtained, and the expression is:

[0130]

[0131] R = det(M) - k × (trace(M)) 2

[0132] where det(M) is the determinant of M, trace(M) is the trace of M, and k is a regulation factor;

[0133] After normalization, R is used as the local regulation factor to obtain the structural response map of the chromatid.

[0134] The module M4 includes:

[0135] Calculate the global mean R of the structural response map mean and the standard deviation R std , divide the gray range of the chromosome image into J merged intervals, and extract the mean value R of the structural tensor response of the corresponding region for each merged interval j , 1 ≤ j ≤ J; according to the deviation between this response value and the global mean R mean of the structural response map, combined with the standard deviation for normalization processing, calculate the weight factor weight of this interval, and limit the weight factor within 0.5 - 2.0. The expression is:

[0136]

[0137] Use this weight factor to adjust the values of each merged interval of the original histogram. If the current merged interval is not the last merged interval, that is, j < J, then according to the change trend of the histogram values, calculate the difference between adjacent merged intervals H[j] and H[j + 1] and obtain the corresponding adjustment value diff. The expression is:

[0138]

[0139] where N is a normalization parameter used to adjust the change amount of each step;

[0140] Then, calculate the histogram values of the front and back parts of each histogram merged interval;

[0141] When H[j] < H[j + 1], keep the histogram values of the center points C j and C j+1 unchanged, increase the histogram value of the j th point <0000;469>to the right of C and decrease the histogram value of the j+1 th point to the left of C , and add each point to the interpolation list l;

[0142] l(C j ) = H[j], l(C j+1 ) = H[j + 1]

[0143] <;

[0144] When H[j] > H[j+1], the center point C of each merged interval is maintained. j C j+1 The histogram values ​​remain unchanged, and C is reduced. j The right side Points Histogram values ​​and C j+1 Left side Points The histogram values ​​are calculated, and each point is added to the interpolation list l;

[0145] l(C j )=H[j],l(C j+1 )=H[j+1]

[0146]

[0147] If the current merged interval is the last interval, i.e., j = J, then directly set the interval center C. j Multiply by the weight factor and add to the interpolation list l;

[0148] A complete 256-level grayscale histogram is generated using linear interpolation. The interpolated histogram is then normalized to ensure that its total number of pixels is consistent with the total number of pixels in the ROI region of the chromosome monomorphic image.

[0149] Those skilled in the art will understand that, in addition to implementing the system, apparatus, and their modules provided by this invention in purely computer-readable program code, the same program can be implemented in the form of logic gates, switches, application-specific integrated circuits, programmable logic controllers, and embedded microcontrollers by logically programming the method steps. Therefore, the system, apparatus, and their modules provided by this invention can be considered a hardware component, and the modules included therein for implementing various programs can also be considered structures within the hardware component; alternatively, modules for implementing various functions can be considered both software programs implementing the method and structures within the hardware component.

[0150] Specific embodiments of the present invention have been described above. It should be understood that the present invention is not limited to the specific embodiments described above, and those skilled in the art can make various changes or modifications within the scope of the claims, which do not affect the essence of the present invention. Unless otherwise specified, the embodiments and features described in this application can be arbitrarily combined with each other.

Claims

1. A self-adapting weight assignment chromosome strip contrast enhancement method, characterized in that, The method comprises the following steps: Step 1: sharpening preprocessing is performed on the original chromosome gray image by using an edge enhancement convolution operator after normalization; Step 2: target detection and instance segmentation of chromosome monomers are performed by using an instance segmentation model in deep learning, so that the monomer ROI region of each chromosome is obtained, including a bounding box and a mask; Step 3: linear stretching is performed on the gray value of the chromosome monomer ROI region, the gray value is mapped to 0 to 255, the horizontal and vertical gradients of the chromosome monomer ROI region are calculated, the structure tensor is constructed, and the structure response value reflecting the texture structure intensity and direction stability of the pixel neighborhood is obtained; Step 4: the gray value of the chromosome monomer ROI region is divided into a plurality of histogram merging intervals, and the structure tensor response mean value of each interval is extracted; Based on the deviation and standard deviation of the mean value and the global response, the weight factor of each merging interval is calculated; based on the weight factor, the value of the original histogram merging interval is adjusted, and finally the complete target histogram is obtained by interpolation; Step 5: the image enhancement of the chromosome monomer ROI region is performed by iterative histogram matching, and the enhancement effect is evaluated by using a contrast improvement ratio CIR index, and the iteration is stopped when the difference value of CIR is lower than a set threshold for two times in succession, and it is determined that the contrast enhancement of the chromosome monomer has reached the expected requirement.

2. The adaptive weight assignment chrominance band contrast enhancement method of claim 1, wherein, The step 1 comprises: an edge enhancement convolution kernel with center enhancement and periphery neutralization is used to enhance the edges and details of the image, and smooth areas are suppressed at the same time; the edge enhancement convolution kernel is used for sharpening operation on the intermediate chromosome gray image, so that the strip structure and edge information of the chromosome monomer are highlighted.

3. The adaptive weight assignment chrominance band contrast enhancement method of claim 1, wherein, The step 2 comprises: the original chromosome gray image is input into an instance segmentation model Mask-RCNN, the model learns local features in the image to identify, locate and distinguish each individual chromosome, and generates corresponding masks, bounding boxes and confidence scores; the mask is used for accurate demarcation of the chromosome region to provide the ROI region for subsequent chromosome contrast; the bounding box is used to represent the position of each chromosome; and the confidence score is used to evaluate the confidence degree of the model to the detection result, and the higher the score is, the greater the probability that the instance is a chromosome is.

4. The adaptive weight assignment chrominance band contrast enhancement method of claim 1, wherein, The step 3 comprises: Linear stretching is performed on the gray value of the chromosome monomer ROI region, and the gray value is mapped to 0 to 255; the gradient I x , y of the horizontal direction and the vertical direction of the image of the chromosome monomer ROI region is calculated respectively xx , yy and the product term I xy of the horizontal and vertical gradients are constructed by pixel-by-pixel combination of the gradient image, that is, the three basic components of the structure tensor, the expression is: I xx = I x 2 I yy = I y 2 I xy = I x x I y Gaussian filtering is adopted to smooth each component of the tensor to obtain a structure tensor M and a structure response value R representing the intensity and direction stability of the texture structure in each pixel neighborhood, and the expression is as follows: R = det(M) - k x (trace(M)) 2 Wherein, det(M) is the determinant of M, trace(M) is the trace of M, and k is an adjustment factor; After normalization, the structure response graph of the chromosome monomer is obtained by taking R as a local adjustment factor.

5. The adaptive weight assignment chrominance band contrast enhancement method of claim 1, wherein, The step 4 comprises: The global mean value R of the structural response map is calculated mean and the standard deviation R std The gray scale range of the chromosome image is divided into J merging intervals, and the structural response mean value R of the corresponding region is extracted for each merging interval j , 1≤j≤J; according to the deviation of the response value from the global mean value R mesn of the structural response map, the weight factor weight of the interval is calculated by normalizing the standard deviation, and the weight factor is limited within 0.5-2.0, and the expression is: The weight factor is used to adjust the value of each merging interval of the original histogram, if the current merging interval is not the last merging interval, that is, j Wherein, N is a normalization parameter for adjusting the change amount of each step; Then, the histogram values of the two parts before and after each histogram merging interval are calculated. When H[j] < H[j + 1], keep the histogram values of the center points C j and C j+1 unchanged, increase the histogram value of the j rightmost point and decrease the histogram value of the j+1 leftmost point , and add each point to the interpolation list l; l(C j ) = H[j], l(C j+1 ) = H[j+1] When H[j] > H[j+1], the center point C of each merged interval is maintained. j C j+1 The histogram values ​​remain unchanged, and C is reduced. j The right side Points Histogram values ​​and C j+1 Left side Points The histogram values ​​are calculated, and each point is added to the interpolation list l; l(C j ) = H[j], l(C j+1 ) = H[j+1] If the current merging interval is the last interval, i.e. j = J, the interval center C j is added to the interpolation list I after being multiplied with the weight factor weight. Using linear interpolation method, generate a complete 256-level gray histogram, normalize the interpolated histogram, ensure that the total pixel number is consistent with the total pixel number of the ROI region of the chromosome monomer image.

6. An adaptive weight assignment chromosome band contrast enhancement system, characterized by, Comprise: Module M1: through edge enhancement convolution operator, normalized and applied to the original chromosome gray image, sharpening preprocessing; Module M2: using instance segmentation model in deep learning to detect and instance segmentation of chromosome monomer target, to obtain the monomer ROI region of each chromosome, including bounding box and mask; Module M3: linear stretching is carried out on the gray value of chromosome monomer ROI region, and the gray value is mapped to 0 to 255; the horizontal and vertical gradients of the chromosome monomer ROI region are calculated, the structure tensor is constructed, and the structure response value reflecting the texture structure intensity and direction stability of the pixel neighborhood is obtained; Module M4: the gray of the chromosome monomer ROI region is divided into multiple histogram merging intervals, and the structure tensor response mean value of each interval is extracted; Based on the deviation and standard deviation of the mean value and the global response, the weight factor of each merging interval is calculated; based on the weight factor, the value of the original histogram merging interval is adjusted, and finally the complete target histogram is obtained by interpolation; Module M5: through iterative histogram matching, the image enhancement of the chromosome monomer ROI region is carried out, and the contrast improvement ratio CIR index is used to evaluate the enhancement effect; when the difference value of CIR is less than the set threshold for two times in succession, the iteration is stopped, and it is determined that the contrast enhancement of the chromosome monomer has reached the expected requirement.

7. The adaptive weight assignment chrominance band contrast enhancement system of claim 6, wherein, The module M1 comprises: using the edge enhancement convolution kernel with center enhancement and periphery neutralization, the edges and details of the image are enhanced, and the smooth area is suppressed at the same time; the edge enhancement convolution kernel performs sharpening operation on the intermediate chromosome gray image, highlights the strip structure and edge information of the chromosome monomer.

8. The adaptive weight allocation chrominance band contrast enhancement system of claim 6, wherein, The module M2 comprises: inputting the original chromosome gray image into the instance segmentation model Mask-RCNN, the model learns the local features in the image to identify, locate and distinguish each individual chromosome, and generates the corresponding mask, bounding box and confidence score; the mask is used for accurate demarcation of the chromosome region to provide the ROI region for subsequent chromosome contrast; the bounding box is used to represent the position of each chromosome; the confidence score is used to evaluate the confidence degree of the model to the detection result, and the higher the score, the greater the probability that the instance is a chromosome.

9. The adaptive weight allocation chrominance band contrast enhancement system of claim 6, wherein, The module M3 comprises: Linear stretching is applied to the grayscale values ​​of the chromosome monosomy ROI region to map them to 0 to 255; the horizontal and vertical gradients I of the chromosome monosomy ROI region image are calculated respectively. x I y The three basic components of the structure tensor are constructed by combining the gradient map pixel by pixel, namely the horizontal gradient squared term I. xx Vertical gradient squared term I yy And the product term I of the horizontal and vertical gradients xy The expression is: I xx = I x 2 I yy = I y 2 I xy = I x x I y Gaussian filtering is adopted to smooth each component of the tensor to obtain the structure tensor M and the structure response value R representing the intensity and direction stability of the texture structure in each pixel neighborhood, and the expression is as follows: R = det(M) - k x (trace(M)) 2 Wherein, det(M) is the determinant of M, trace(M) is the trace of M, and k is the adjustment factor; After normalization, the structure response graph of the chromosome monomer is obtained by taking R as the local adjustment factor.

10. The adaptive weight allocation chrominance band contrast enhancement system of claim 6, wherein, The module M4 comprises: The global mean value R of the structural response map is calculated mean and the standard deviation R std The gray scale range of the chromosome image is divided into J merging intervals, and the structural response mean value R of the corresponding region is extracted for each merging interval j , 1≤j≤J; according to the deviation of the response value from the global mean value R mean of the structural response map, the weight factor weight of the interval is calculated by normalizing the standard deviation, and the weight factor is limited within 0.5-2.0, and the expression is: The weight factor is used to adjust the value of each merging interval of the original histogram, if the current merging interval is not the last merging interval, that is, j Wherein, N is a normalization parameter, used to adjust the change of each step; Then, the histogram values of the two parts before and after the merging interval are calculated; When H[j] < H[j + 1], keep the histogram values of the center points C j and C j+1 unchanged, increase the histogram value of the j right -hand point and decrease the histogram value of the j+1 left -hand point , and add each point to the interpolation list l; l(C j ) = H[j], l(C j+1 ) = H[j+1] When H[j] > H[j+1], the center point C of each merged interval is maintained. j C j+1 The histogram values ​​remain unchanged, and C is reduced. j The right side Points Histogram values ​​and C j+1 Left side Points The histogram values ​​are calculated, and each point is added to the interpolation list l; l(C j ) = H[j], l(C j+1 ) = H[j+1] If the current merging interval is the last interval, i.e. j = J, the interval center C j is added to the interpolation list I after being multiplied with the weight factor weight. The linear interpolation method is used to generate a complete 256-level grayscale histogram, and the interpolated histogram is normalized to ensure that the total pixel number is consistent with the total pixel number of the ROI region of the chromosome monomer image.

Images