Adaptive infrared blood vessel imaging enhancement method and device based on image entropy feedback

Abstract

This application provides an adaptive infrared vascular imaging enhancement method and apparatus based on image entropy feedback. It achieves precise adaptive adjustment of infrared emission power through local information entropy calculation and a proportional-integral-differential controller. An enhancement mechanism is constructed, combining multi-scale Gaussian smoothing and Hessian matrix eigenvalue analysis to establish a reliable vascular response extraction strategy. Structural analysis optimization is introduced, iteratively refining skeleton extraction and cross-point calculation to ensure continuous improvement in vascular cross-point localization. This method effectively addresses the shortcomings of traditional techniques in exposure adaptation, vascular enhancement, and topology analysis, providing technical support for infrared vascular imaging-assisted diagnosis.

Description

Technical Field

[0001] This application relates to the field of image processing, specifically to an adaptive infrared vascular imaging enhancement method and apparatus based on image entropy feedback. Background Technology

[0002] Existing infrared vascular imaging enhancement methods have significant shortcomings. Traditional systems perform poorly in image acquisition and exposure control, failing to effectively achieve adaptive adjustment of infrared emission power, thus affecting image quality.

[0003] Furthermore, existing technologies face bottlenecks in vascular feature extraction and enhancement. Most systems lack a robust multi-scale Hessian matrix analysis mechanism and tubular feature discrimination strategy, resulting in less than ideal accuracy in vascular response recognition.

[0004] Existing systems have technical limitations in vascular structure analysis. They lack in-depth analysis of skeleton refinement and intersection detection, making it difficult to achieve efficient vascular topology localization through intersection number calculation. Solving these problems is crucial for improving infrared vascular imaging capabilities. Summary of the Invention

[0005] To address the problems in the existing technology, this application provides an adaptive infrared vascular imaging enhancement method and apparatus based on image entropy feedback, which can effectively solve the shortcomings of traditional technologies in exposure adaptation, vascular enhancement and topology resolution, and provide technical support for infrared vascular imaging.

[0006] To solve at least one of the above problems, this application provides the following technical solution: In a first aspect, this application provides an adaptive infrared vascular imaging enhancement method based on image entropy feedback, comprising: The original infrared image of the face region is acquired by an infrared image acquisition device and divided into several sub-regions according to a preset grid size. The mean gray value and gray level probability distribution of each sub-region are calculated and substituted into the information entropy formula to obtain the local information entropy. The mean gray value of the pixels and the local information entropy are weighted and summed according to a preset weight coefficient to obtain the exposure score. The exposure score is compared with the preset ideal imaging range to calculate the exposure deviation value. The exposure deviation value is used as an error signal input to the incremental proportional-integral-derivative controller to calculate the power adjustment amount. The power adjustment amount is superimposed on the current infrared emission power and sent to the infrared emission tube drive circuit to perform power update. A multi-scale Gaussian smooth space is constructed for the updated infrared image and the Hessian matrix and eigenvalues ​​are calculated at each scale. The eigenvalues ​​are substituted into the tubular feature discrimination formula to calculate the blood vessel response value at each scale and the maximum value is taken in the scale dimension to obtain the blood vessel enhancement image. Adaptive threshold binarization is performed on the enhanced blood vessel image to obtain a binary blood vessel mask. The binary blood vessel mask is input into an iterative thinning algorithm to strip edge pixels and obtain a single-pixel width skeleton map. The eight-neighborhood states of each non-zero pixel in the single-pixel width skeleton map are traversed and the number of intersections is calculated. The coordinates of pixels with an intersection number greater than a preset connection threshold are extracted as a set of blood vessel intersection coordinates and rendered as a highlight mark layer, which is then superimposed on the display terminal for output.

[0007] Furthermore, it also includes: scanning and acquiring the face area using an infrared image acquisition device to obtain an original infrared image; uniformly dividing the original infrared image into several sub-regions in the horizontal and vertical directions according to a preset grid size; summing the pixel grayscale values ​​in each sub-region and dividing by the total number of pixels in the sub-region to obtain the pixel grayscale mean; and counting the number of pixels appearing at each grayscale level in each sub-region and dividing by the total number of pixels in the sub-region to obtain the grayscale probability distribution. Read the probability value corresponding to each gray level in the gray level probability distribution, take the logarithm of each probability value and multiply it by the probability value to obtain the entropy contribution of each gray level, sum the entropy contributions of each gray level and take the negative value to obtain the local information entropy, and associate the pixel gray mean with the local information entropy to form sub-region feature data for subsequent exposure score calculation.

[0008] Furthermore, it also includes: reading the pixel grayscale mean and local information entropy from the sub-region feature data, weighting the pixel grayscale mean and the local information entropy according to a preset weight coefficient and summing them to obtain an exposure score, wherein the preset weight coefficient is composed of brightness weight and feature richness weight and the sum of the two is a unit value; The target grayscale expectation value is read from the preset ideal imaging range. The exposure score is subtracted from the target grayscale expectation value to obtain the exposure deviation value. The exposure deviation value is output to the power adjustment stage as an error signal for the incremental proportional-integral-derivative controller to call.

[0009] Furthermore, it also includes: inputting the exposure deviation value as the current time error signal into the incremental proportional-integral-derivative controller, reading the current time error signal, the previous time error signal, and the error signals of the two previous times, calculating the proportional increment, integral increment, and derivative increment according to the preset proportional coefficient, integral coefficient, and derivative coefficient, and summing them to obtain the power adjustment amount; The current infrared emission power is read and the power adjustment amount is added to the current infrared emission power to obtain the target emission power. The target emission power is compared with the preset power upper limit value and the preset power lower limit value and constrained to the allowable range to obtain the constrained emission power. The constrained emission power is sent to the infrared emitting tube drive circuit to perform power update.

[0010] Furthermore, it also includes: performing Gaussian convolution smoothing on the updated infrared images according to a preset scale parameter sequence to obtain a multi-scale Gaussian smoothing space; calculating the horizontal second-order partial derivative, the vertical second-order partial derivative, and the mixed second-order partial derivative for each pixel in the multi-scale Gaussian smoothing space and assembling them into a Hessian matrix; and solving the characteristic equation of the Hessian matrix to obtain the first eigenvalue and the second eigenvalue. Based on the first feature value and the second feature value, a speckled discriminant factor and a structural sensitivity factor are calculated. The speckled discriminant factor and the structural sensitivity factor are substituted into the tubular feature discriminant formula to calculate the vascular response value at each scale. The maximum value of each pixel is extracted along the scale dimension from the vascular response values ​​at each scale to obtain the vascular enhancement image.

[0011] Furthermore, it also includes: dividing the enhanced vascular image into local regions according to a preset neighborhood window size, statistically analyzing the pixel grayscale distribution in each local region and calculating a local threshold, comparing the grayscale value of each pixel with the corresponding local threshold, and marking pixels greater than the local threshold as foreground and pixels less than the local threshold as background to obtain a binary vascular mask; The binary blood vessel mask is input into an iterative refinement algorithm. In each iteration, the foreground pixels are traversed and it is determined whether the pixel is a deletable edge pixel based on the eight-neighbor connectivity condition. Edge pixels that meet the deletion condition are set as background and pixels that do not meet the deletion condition are retained. The iteration is repeated until there are no deletable pixels to obtain a single-pixel width skeleton map.

[0012] Furthermore, it also includes: traversing each non-zero pixel in the single-pixel width skeleton diagram, reading the pixel value sequence of the eight neighboring positions around the current pixel, calculating the absolute value of the difference between adjacent pixel values ​​in a clockwise direction, summing them and dividing by two to obtain the cross number, and storing the cross number in association with the coordinates of the current pixel to form a pixel topology attribute record. The set of blood vessel intersection coordinates is obtained by filtering pixel coordinates with a number of intersections greater than a preset connection threshold from the pixel topology attribute record. Highlighted marker graphic elements are generated according to the coordinate positions of each coordinate in the set of blood vessel intersection coordinates and assembled into a highlighted marker layer. The highlighted marker layer is aligned with the current imaging screen and superimposed and rendered to the display terminal for output.

[0013] Secondly, this application provides an adaptive infrared vascular imaging enhancement device based on image entropy feedback, comprising: The pixel calculation module is used to acquire the original infrared image of the face area through the infrared image acquisition device and divide it into several sub-regions according to the preset grid size. It calculates the pixel gray mean and gray level probability distribution for each sub-region and substitutes them into the information entropy formula to obtain the local information entropy. It weights and sums the pixel gray mean and the local information entropy according to the preset weight coefficient to obtain the exposure score. It compares the exposure score with the preset ideal imaging range to calculate the exposure deviation value. The feature processing module is used to input the exposure deviation value as an error signal into the incremental proportional-integral-derivative controller to calculate the power adjustment amount, superimpose the power adjustment amount onto the current infrared emission power and send it to the infrared emitting tube drive circuit to perform power update, construct a multi-scale Gaussian smooth space for the updated infrared image and calculate the Hessian matrix and eigenvalues ​​at each scale, substitute the eigenvalues ​​into the tubular feature discrimination formula to calculate the blood vessel response value at each scale and take the maximum value in the scale dimension to obtain the blood vessel enhancement image; The infrared imaging module is used to perform adaptive threshold binarization on the enhanced blood vessel image to obtain a binary blood vessel mask. The binary blood vessel mask is input into an iterative thinning algorithm to strip edge pixels to obtain a single-pixel width skeleton map. The eight-neighborhood states of each non-zero pixel in the single-pixel width skeleton map are traversed and the number of intersections is calculated. The coordinates of pixels with an intersection number greater than a preset connection threshold are extracted as a set of blood vessel intersection coordinates and rendered as a highlight mark layer, which is then superimposed on the display terminal for output.

[0014] Thirdly, this application provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the steps of the adaptive infrared vascular imaging enhancement method based on image entropy feedback.

[0015] Fourthly, this application provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the adaptive infrared vascular imaging enhancement method based on image entropy feedback.

[0016] Fifthly, this application provides a computer program product, including a computer program / instructions that, when executed by a processor, implement the steps of the adaptive infrared vascular imaging enhancement method based on image entropy feedback.

[0017] As described above, this application provides an adaptive infrared vascular imaging enhancement method and apparatus based on image entropy feedback. Through local information entropy calculation and a proportional-integral-differential controller, precise adaptive adjustment of infrared emission power is achieved. An enhancement mechanism is constructed, combining multi-scale Gaussian smoothing and Hessian matrix eigenvalue analysis to establish a reliable vascular response extraction strategy. Structural analysis optimization is introduced, and continuous improvement in vascular intersection point localization is ensured through iterative refinement of skeleton extraction and intersection number calculation. This method effectively addresses the shortcomings of traditional techniques in exposure adaptation, vascular enhancement, and topology analysis, providing technical support for infrared vascular imaging-assisted diagnosis. Attached Figure Description

[0018] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0019] Figure 1 This is a flowchart illustrating the adaptive infrared vascular imaging enhancement method based on image entropy feedback in an embodiment of this application. Figure 2 This is a structural diagram of the adaptive infrared vascular imaging enhancement device based on image entropy feedback in an embodiment of this application. Detailed Implementation

[0020] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0021] The acquisition, storage, use, and processing of data in this application comply with relevant laws and regulations.

[0022] To address the shortcomings of existing technologies, this application provides an adaptive infrared vascular imaging enhancement method and apparatus based on image entropy feedback. Through local information entropy calculation and a proportional-integral-differential controller, precise adaptive adjustment of infrared emission power is achieved. An enhancement mechanism is constructed, combining multi-scale Gaussian smoothing and Hessian matrix eigenvalue analysis to establish a reliable vascular response extraction strategy. Structural analytical optimization is introduced, iteratively refining skeleton extraction and cross-point calculation to ensure continuous improvement in vascular cross-point localization. This method effectively solves the deficiencies of traditional techniques in exposure adaptation, vascular enhancement, and topology analysis, providing technical support for infrared vascular imaging-assisted diagnosis.

[0023] To effectively address the shortcomings of traditional techniques in areas such as exposure adaptation, vessel enhancement, and topology resolution, and to provide technical support for infrared vascular imaging, this application provides an embodiment of an adaptive infrared vascular imaging enhancement method based on image entropy feedback. See [link to embodiment]. Figure 1 The adaptive infrared vascular imaging enhancement method based on image entropy feedback specifically includes the following: Step S101: Acquire the original infrared image of the face area through the infrared image acquisition device and divide it into several sub-regions according to the preset grid size. Calculate the pixel grayscale mean and grayscale probability distribution for each sub-region and substitute them into the information entropy formula to obtain the local information entropy. Weight the pixel grayscale mean and the local information entropy according to the preset weight coefficient to obtain the exposure score. Compare the exposure score with the preset ideal imaging range to calculate the exposure deviation value. This embodiment uses an infrared image acquisition device to scan and acquire raw infrared images of the facial area. The infrared image acquisition device includes a near-infrared light source and a complementary metal-oxide-semiconductor (CMOS) sensor. The near-infrared light source emits infrared light with a wavelength within a preset range towards the facial area. The CMOS sensor receives the infrared light reflected and scattered by the skin tissue and converts it into a digital image signal. The raw infrared image is stored in the form of a two-dimensional pixel array, with each pixel position recording the infrared light intensity information of the corresponding spatial point.

[0024] After the original infrared image is acquired, this embodiment divides it into several sub-regions in both the horizontal and vertical directions according to a preset grid size. The preset grid size defines the number of blocks in the horizontal and vertical directions. In this embodiment, the original infrared image is divided into regularly arranged rectangular sub-regions according to this number of blocks. Each sub-region covers different anatomical locations of the face, with the forehead region, periorbital region, cheek region, and bridge of the nose region falling within their respective sub-regions.

[0025] Accordingly, this embodiment calculates the average pixel grayscale value for each sub-region. Specifically, this embodiment traverses all pixels within a single sub-region and reads the grayscale value of each pixel. The grayscale values ​​are then summed and divided by the total number of pixels contained in the sub-region to obtain the average pixel grayscale value. The average pixel grayscale value reflects the overall brightness level of the sub-region. The forehead region, due to its thinner skin and stronger reflection, typically exhibits a higher average pixel grayscale value, while the cheek region, due to its deeper tissue and greater absorption, typically exhibits a lower average pixel grayscale value.

[0026] After the pixel grayscale mean is calculated, this embodiment statistically analyzes the grayscale probability distribution of each sub-region and substitutes it into the information entropy formula to obtain the local information entropy. The statistical process of grayscale probability distribution involves traversing each pixel within the sub-region and recording the number of pixels appearing at each grayscale level. The number of pixels at each grayscale level is divided by the total number of pixels in the sub-region to obtain the probability value corresponding to each grayscale level. In this embodiment, the probability value is read, and the logarithm of each probability value is taken to base 2 and multiplied by the probability value to obtain the entropy contribution of each grayscale level. The entropy contributions of all grayscale levels are summed and the negative value is taken to obtain the local information entropy. The local information entropy reflects the dispersion of the grayscale distribution within the sub-region. Regions rich in vascular features exhibit higher local information entropy due to the diversity of grayscale levels.

[0027] Based on the aforementioned pixel grayscale mean and local information entropy, this embodiment calculates the exposure score by weighting the two together according to a preset weighting coefficient. The exposure score can be expressed as: Q = w1 U + w2 R.

[0028] In the formula, Q is the exposure score of the current sub-region; U is the average pixel grayscale value calculated above; R is the local information entropy calculated above; w1 is the brightness weight, w2 is the feature richness weight, and the sum of the two is a unit value and both are non-negative. The Q comprehensively considers the brightness level and feature richness of the quantum region, providing a fusion evaluation index for subsequent power adjustment.

[0029] After the exposure score is calculated, this embodiment compares it with a preset ideal imaging range to obtain an exposure deviation value. The preset ideal imaging range includes the target grayscale expectation value. This embodiment calculates the exposure deviation value by subtracting the exposure score Q from the target grayscale expectation value. A positive exposure deviation value indicates that the current sub-region is overexposed, and a negative exposure deviation value indicates that the current sub-region is underexposed. The exposure deviation value is output as an error signal to the incremental proportional-integral-derivative controller in the subsequent step S102 to drive the closed-loop adjustment of the infrared emission power.

[0030] Step S102: The exposure deviation value is used as an error signal input to the incremental proportional-integral-derivative controller to calculate the power adjustment amount. The power adjustment amount is superimposed on the current infrared emission power and sent to the infrared emitting tube drive circuit to perform power update. A multi-scale Gaussian smooth space is constructed for the infrared image after the update, and the Hessian matrix and eigenvalues ​​are calculated at each scale. The eigenvalues ​​are substituted into the tubular feature discrimination formula to calculate the blood vessel response value at each scale, and the maximum value is taken in the scale dimension to obtain the blood vessel enhancement image. In this embodiment, the exposure deviation value generated in step S101 is read and used as the current error signal input to the incremental proportional-integral-derivative (PID) controller. The IPD controller maintains a timing record of the error signals. In this embodiment, while receiving the current error signal, the error signals from the previous two time steps are read simultaneously. The error signals from these three time steps jointly participate in the calculation of the power adjustment. The incremental structure ensures that the controller outputs the power change rather than the absolute power value, avoiding adjustment instability caused by integral saturation.

[0031] After the error signal is read, this embodiment calculates the proportional increment, integral increment, and differential increment based on preset proportional, integral, and differential coefficients, respectively. The proportional increment is obtained by multiplying the difference between the error signal at the current moment and the error signal at the previous moment by the proportional coefficient, reflecting the immediate response to error changes. The integral increment is obtained by multiplying the error signal at the current moment by the integral coefficient, used to eliminate steady-state deviation. The differential increment is obtained by subtracting twice the error signal at the previous moment from the error signal at the current moment, adding the error signals from two moments ago, and then multiplying by the differential coefficient, used to suppress the oscillating trend of error changes. This embodiment sums the proportional increment, integral increment, and differential increment to obtain the power adjustment amount.

[0032] Accordingly, this embodiment reads the current infrared emission power and adds it to the power adjustment amount to obtain the target emission power. To prevent power exceeding the limit from damaging the infrared emitting tube or causing image quality degradation, this embodiment compares the target emission power with preset power upper limit and preset power lower limit values. If the target emission power exceeds the preset power upper limit value, it is constrained to the preset power upper limit value; if the target emission power is lower than the preset power lower limit value, it is constrained to the preset power lower limit value. The constrained emission power is within the allowable range. This embodiment sends the constrained emission power to the infrared emitting tube drive circuit to perform power update. The drive circuit adjusts the drive current of the infrared emitting tube according to the received power value.

[0033] After the power update is performed, this embodiment constructs a multi-scale Gaussian smoothing space for the updated infrared image. Specifically, this embodiment presets a sequence of scale parameters, which contains several increasing scale factors. Smaller scale factors correspond to narrower blood vessel widths, and larger scale factors correspond to wider blood vessel widths. This embodiment performs Gaussian convolution smoothing on the updated infrared image sequentially according to the scale parameter sequence, and the smoothing results corresponding to each scale factor form a multi-scale Gaussian smoothing space.

[0034] Based on the aforementioned multi-scale Gaussian smoothing space, this embodiment calculates the Hessian matrix for each pixel in the image at each scale. The Hessian matrix is ​​assembled from the second-order partial derivatives in the horizontal direction, the second-order partial derivatives in the vertical direction, and a mixture of second-order partial derivatives, describing the local curvature characteristics of the gray-level surface at that pixel. In this embodiment, the characteristic equation of the Hessian matrix is ​​solved to obtain the first eigenvalue and the second eigenvalue, and they are sorted by absolute value so that the absolute value of the first eigenvalue is not greater than the absolute value of the second eigenvalue.

[0035] After the eigenvalues ​​are calculated, this embodiment calculates a speckle discrimination factor and a structural sensitivity factor based on the first and second eigenvalues. The speckle discrimination factor is obtained by dividing the absolute value of the first eigenvalue by the absolute value of the second eigenvalue, and is used to distinguish tubular structures from circular specks. The structural sensitivity factor is obtained by taking the square root of the sum of the squares of the two eigenvalues, and is used to suppress low-contrast background noise. In this embodiment, the speckle discrimination factor and the structural sensitivity factor are substituted into the tubular feature discrimination formula to calculate the vascular response values ​​at various scales. The tubular feature discrimination formula can be expressed as: F = exp(-A A / (2 c1 c1)) (1 - exp(-B B / (2 c2 c2))).

[0036] In the formula, F is the blood vessel response value of the current pixel at the current scale; A is the speckle discrimination factor calculated above; B is the structure sensitivity factor calculated above; c1 and c2 are sensitivity threshold parameters, which are determined according to the blood vessel imaging characteristics during the system calibration stage.

[0037] After the vascular response values ​​at each scale are calculated, this embodiment extracts the maximum value of each pixel along the scale dimension to obtain a vascular enhancement image. Each pixel value in the vascular enhancement image represents the confidence level of the presence of a tubular vascular structure at that location; vascular regions exhibit higher response values, while non-vascular regions exhibit lower response values. The vascular enhancement image is output to subsequent step S103 for adaptive threshold binarization and skeleton extraction processing.

[0038] Step S103: Perform adaptive threshold binarization on the enhanced blood vessel image to obtain a binary blood vessel mask. Input the binary blood vessel mask into an iterative thinning algorithm to strip edge pixels and obtain a single-pixel width skeleton map. Traverse the eight-neighborhood states of each non-zero pixel in the single-pixel width skeleton map and calculate the number of intersections. Extract the coordinates of pixels with an intersection number greater than a preset connection threshold as a set of blood vessel intersection coordinates and render them as a highlight mark layer, which is then superimposed on the display terminal for output.

[0039] This embodiment reads the enhanced vascular image generated in step S102 and performs adaptive threshold binarization on it. Adaptive threshold binarization divides the enhanced vascular image into several local regions according to a preset neighborhood window size, which is determined based on the vascular width range and image resolution. In this embodiment, the pixel grayscale distribution is statistically analyzed within each local region, and a local threshold is calculated. This local threshold is adaptively generated based on the statistical characteristics of the pixel grayscale within that region, and different regions have different local thresholds due to differences in tissue background.

[0040] After the local threshold calculation is completed, this embodiment compares the grayscale value of each pixel with the corresponding local threshold to separate the foreground and background. Pixels with grayscale values ​​greater than the local threshold are marked as foreground, and pixels with grayscale values ​​less than the local threshold are marked as background. Foreground pixels correspond to vascular regions, while background pixels correspond to non-vascular regions. This embodiment assembles the marking results into a binary vascular mask, which stores the foreground or background attributes of each pixel in binary form.

[0041] Accordingly, this embodiment uses an iterative thinning algorithm to strip edge pixels from the binary blood vessel mask. In each iteration, the iterative thinning algorithm traverses all foreground pixels in the binary blood vessel mask and analyzes the eight-neighbor connectivity state of each foreground pixel. This embodiment determines whether the current foreground pixel is a removable edge pixel based on the eight-neighbor connectivity condition. This connectivity condition includes constraints on the number of neighboring foreground pixels and connectivity preservation constraints. Edge pixels that satisfy the deletion condition are set as background in this iteration.

[0042] After a single iteration, this embodiment checks whether any edge pixels were deleted in that iteration. If deleted pixels exist, the next iteration continues; otherwise, the iteration terminates. The iterative thinning algorithm gradually shrinks the blood vessel region to the centerline position by peeling away edge pixels layer by layer. At the end of the iteration, the remaining foreground pixels form a single-pixel width skeleton map. This single-pixel width skeleton map strips away the width information of the original blood vessel image, retaining only the topological orientation information of the blood vessel.

[0043] Based on the aforementioned single-pixel width skeleton diagram, this embodiment traverses each non-zero pixel and calculates the crossover number. For the current non-zero pixel, this embodiment reads the pixel value sequence of its eight neighboring positions, which are arranged in a clockwise direction to form a closed loop. This embodiment calculates the absolute value of the difference between adjacent pixel values ​​in a clockwise direction, sums all differences, and divides by two to obtain the crossover number. The crossover number can be expressed as: N = (|p0 - p1| + |p1 - p2| + |p2 - p3| + |p3 - p4| + |p4 - p5| + |p5- p6| + |p6 - p7| + |p7 - p0|) / 2.

[0044] In the formula, N is the number of intersections of the current pixel; p0 to p7 are the values ​​of the eight neighboring pixels arranged clockwise, and their values ​​are either zero or one. In this embodiment, the number of intersections is associated with the coordinates of the current pixel to form a pixel topology attribute record.

[0045] After the pixel topology attribute record is generated, this embodiment filters pixel coordinates with a crossover number greater than a preset connection threshold. The preset connection threshold distinguishes between ordinary connection points and crossover / bifaction points. A crossover number of one indicates the pixel is an endpoint, a crossover number of two indicates the pixel is an ordinary connection point, and a crossover number greater than the preset connection threshold indicates the pixel is a crossover point or bifurcation point. This embodiment extracts the pixel coordinates that meet the conditions to obtain a set of vascular crossover point coordinates, which records key locations where multidirectional blood flow converges in the facial vascular network.

[0046] After the set of coordinates of the blood vessel intersections is generated, this embodiment generates highlighted graphic elements based on each coordinate position. These highlighted graphic elements use preset geometric shapes and color configurations. This embodiment assembles the rendered graphic elements at each intersection coordinate into a highlighted marker layer. The highlighted marker layer is then aligned with the current imaging screen. This alignment process uses the mapping relationship between the image coordinate system and the display coordinate system to complete spatial position matching. This embodiment overlays and renders the aligned highlighted marker layer onto the display terminal for output, allowing operators to identify blood vessel intersections and avoid corresponding areas when performing facial operations.

[0047] As described above, the adaptive infrared vascular imaging enhancement method based on image entropy feedback provided in this application can achieve precise adaptive adjustment of infrared emission power through local information entropy calculation and a proportional-integral-differential controller. An enhancement mechanism is constructed, combining multi-scale Gaussian smoothing and Hessian matrix eigenvalue analysis to establish a reliable vascular response extraction strategy. Structural analysis optimization is introduced, and continuous improvement in vascular intersection point localization is ensured through iterative refinement of skeleton extraction and intersection number calculation. This method effectively addresses the shortcomings of traditional techniques in exposure adaptation, vascular enhancement, and topology analysis, providing technical support for infrared vascular imaging-assisted diagnosis.

[0048] In one embodiment of the adaptive infrared blood vessel imaging enhancement method based on image entropy feedback in this application, it may further include the following: Step S201: Scan the face area with an infrared image acquisition device to obtain an original infrared image. Divide the original infrared image into several sub-regions in the horizontal and vertical directions according to a preset grid size. Sum the pixel gray values ​​in each sub-region and divide by the total number of pixels in the sub-region to obtain the pixel gray average. Count the number of pixels at each gray level in each sub-region and divide by the total number of pixels in the sub-region to obtain the gray level probability distribution. Step S202: Read the probability value corresponding to each gray level in the gray level probability distribution, take the logarithm of each probability value and multiply it with the probability value to obtain the entropy contribution of each gray level, sum the entropy contributions of each gray level and take the negative value to obtain the local information entropy, and associate and store the pixel gray mean with the local information entropy to form sub-region feature data for subsequent exposure score calculation.

[0049] In this embodiment, an infrared image acquisition device scans and acquires the original infrared image by scanning the facial area. After activation, the near-infrared light source of the infrared image acquisition device continuously emits infrared light towards the facial area. This infrared light is reflected and scattered by the skin surface and subcutaneous tissue and then received by a complementary metal-oxide-semiconductor (CMOS) sensor. The CMOS sensor converts the received light signal into an electrical signal and performs analog-to-digital conversion to generate digital image data. This digital image data is organized into a two-dimensional pixel array to form the original infrared image.

[0050] After the original infrared image is acquired, this embodiment divides it into several sub-regions in both the horizontal and vertical directions according to a preset grid size. The preset grid size is defined in the form of the number of horizontal and vertical blocks. This embodiment reads the number of blocks and calculates the pixel span of each sub-region in both the horizontal and vertical directions. The segmentation process starts from the upper left corner of the original infrared image and delineates the boundary range of each sub-region sequentially according to the pixel span. All sub-regions cover the entire area of ​​the original infrared image, and there is no overlap between adjacent sub-regions.

[0051] Accordingly, this embodiment sums the pixel grayscale values ​​within each sub-region and divides them by the total number of pixels in the sub-region to obtain the average pixel grayscale value. Specifically, this embodiment traverses all pixel positions within the boundary range of a single sub-region, reads the grayscale value stored at each pixel position, and accumulates it into a total grayscale value variable. After traversal, this embodiment counts the total number of pixels contained in the sub-region and divides the sum of grayscale values ​​by the total number of pixels to obtain the average pixel grayscale value. The average pixel grayscale value is stored in floating-point form, reflecting the average brightness level of the sub-region.

[0052] After the pixel grayscale mean is calculated, this embodiment counts the number of pixels at each grayscale level within each sub-region and divides this number by the total number of pixels in the sub-region to obtain the grayscale level probability distribution. This embodiment initializes a grayscale level counting array, the length of which corresponds to the total number of grayscale levels. When traversing each pixel within a sub-region, this embodiment reads the grayscale value of the current pixel and increments the count value at the corresponding position in the grayscale level counting array. After traversal, this embodiment divides each count value in the grayscale level counting array by the total number of pixels in the sub-region to obtain the probability value corresponding to each grayscale level and forms the grayscale level probability distribution.

[0053] Based on the aforementioned gray-level probability distribution, this embodiment reads the probability value corresponding to each gray level for local information entropy calculation. This embodiment iterates through each probability value in the gray-level probability distribution. For gray levels with a probability value greater than zero, the base-2 logarithm of the probability value is taken and multiplied by the logarithm to obtain the entropy contribution of that gray level. For gray levels with a probability value equal to zero, this embodiment sets its entropy contribution to zero to avoid numerical anomalies in logarithmic operations. The entropy contribution reflects the degree to which each gray level contributes to the overall information uncertainty.

[0054] After calculating the entropy contribution of each gray level, this embodiment sums them and takes the negative value to obtain the local information entropy. The summation process accumulates the entropy contributions of all gray levels. Since each entropy contribution is a non-positive value, which is the product of a probability value and a logarithm, the accumulated result is also non-positive. This embodiment takes the negative value of the accumulated result to obtain a non-negative local information entropy value. The larger the local information entropy value, the more dispersed the gray level distribution in that sub-region, and the richer the gray level between blood vessels and tissues.

[0055] After the local information entropy calculation is completed, this embodiment associates and stores the pixel grayscale mean with the local information entropy to form sub-region feature data. The sub-region feature data uses the sub-region index as the key, organizing the pixel grayscale mean and local information entropy corresponding to that sub-region into data records. This embodiment writes the feature data of all sub-regions into a sub-region feature data set, which is then read by subsequent step S203 for exposure score calculation and exposure deviation value generation.

[0056] In one embodiment of the adaptive infrared blood vessel imaging enhancement method based on image entropy feedback in this application, it may further include the following: Step S301: Read the pixel grayscale mean and local information entropy from the sub-region feature data, and weight the pixel grayscale mean and the local information entropy according to the preset weight coefficient and sum them to obtain the exposure score. The preset weight coefficient consists of brightness weight and feature richness weight, and the sum of the two is a unit value. Step S302: Read the target grayscale expectation value from the preset ideal imaging range, subtract the exposure score from the target grayscale expectation value to obtain the exposure deviation value, and output the exposure deviation value to the power adjustment stage as an error signal for the incremental proportional-integral-derivative controller to call.

[0057] This embodiment reads the sub-region feature data generated in step S202 above and extracts the pixel grayscale mean and local information entropy from it. The sub-region feature data is organized and stored using the sub-region index as the key. In this embodiment, the data records corresponding to each sub-region are traversed in index order, and the pixel grayscale mean and local information entropy values ​​of each sub-region are read sequentially. The pixel grayscale mean reflects the brightness level of the sub-region, and the local information entropy reflects the dispersion and feature richness of the grayscale distribution within the sub-region.

[0058] After the pixel grayscale mean and local information entropy are read, this embodiment weights them according to preset weighting coefficients. The preset weighting coefficients consist of a brightness weight and a feature richness weight. The brightness weight adjusts the contribution of the pixel grayscale mean to the exposure score, and the feature richness weight adjusts the contribution of the local information entropy to the exposure score. Both the brightness weight and the feature richness weight are non-negative values, and their sum is a unit value. This constraint ensures that the exposure score remains within a reasonable numerical range.

[0059] Accordingly, this embodiment obtains the exposure score by summing the weighted average pixel grayscale value and the weighted local information entropy. The exposure score can be expressed as: D = k1 J + k2 T.

[0060] In the formula, D is the exposure score of the current sub-region; J is the average pixel grayscale value read above; T is the local information entropy read above; k1 is the brightness weight, k2 is the feature richness weight, and the sum of the two is a unit value and both are non-negative. The D comprehensively reflects the brightness state and feature presentation quality of the sub-region, providing a fusion evaluation benchmark for subsequent exposure deviation calculation.

[0061] After the exposure score calculation is completed, this embodiment reads the target grayscale expectation value from the preset ideal imaging range. The preset ideal imaging range defines the ideal exposure range for vascular imaging, and the target grayscale expectation value is located at the center of this range, representing the exposure score reference value when the contrast between blood vessels and tissues is ideal. This embodiment loads the target grayscale expectation value into memory for use in exposure deviation calculation.

[0062] Based on the aforementioned exposure score D and the target grayscale expectation value, this embodiment calculates the difference between the two to obtain the exposure deviation value. The exposure deviation value is calculated by subtracting the target grayscale expectation value from the exposure score; the difference reflects the degree of deviation between the actual exposure state and the ideal exposure state of the current sub-region. A positive exposure deviation value indicates that the current sub-region is overexposed, and the infrared emission power needs to be reduced to avoid overexposure. A negative exposure deviation value indicates that the current sub-region is underexposed, and the infrared emission power needs to be increased to enhance subcutaneous penetration depth.

[0063] After the exposure deviation value is calculated, this embodiment outputs it to the power adjustment stage as an error signal. The exposure deviation value is written into the error signal buffer in numerical form for subsequent reading by the incremental proportional-integral-derivative (PID) controller in step S401. The IPD controller calculates the power adjustment amount based on the error signal and drives the closed-loop adjustment of the infrared emission power to make the exposure score approach the target grayscale expectation value.

[0064] In one embodiment of the adaptive infrared blood vessel imaging enhancement method based on image entropy feedback in this application, it may further include the following: Step S401: Input the exposure deviation value as the current time error signal into the incremental proportional-integral-derivative controller, read the current time error signal, the previous time error signal, and the error signals of the two previous times, calculate the proportional increment, integral increment, and derivative increment according to the preset proportional coefficient, integral coefficient, and derivative coefficient, and sum them to obtain the power adjustment amount; Step S402: Read the current infrared emission power and add the power adjustment amount to the current infrared emission power to obtain the target emission power. Compare the target emission power with the preset power upper limit value and the preset power lower limit value and constrain it to the allowable range to obtain the constrained emission power. Send the constrained emission power to the infrared emitting tube drive circuit to perform power update.

[0065] This embodiment reads the exposure deviation value generated in step S302 and inputs it as the current error signal into the incremental proportional-integral-derivative (PID) controller. The IPD controller uses an incremental output form, where the output is the power change rather than the absolute power value. This structure avoids the saturation problem that may occur during the integral term accumulation process in traditional position-type controllers. This embodiment writes the current error signal to the current position in the error signal timing queue, which maintains error signal records for the three most recent moments.

[0066] After the error signal at the current moment is written, this embodiment reads the error signal at the current moment, the error signal at the previous moment, and the error signal at the two moments before that from the error signal timing queue. The error signals at these three moments correspond to the exposure deviation status of the most recent three sampling periods, and this embodiment loads them into the calculation buffer for subsequent incremental calculation. If the error signal timing queue is not full at the initial stage of system startup, this embodiment fills the missing error signals with zero values ​​to ensure the continuity of the calculation process.

[0067] Accordingly, this embodiment calculates the proportional increment based on a preset proportional coefficient. The proportional increment is obtained by multiplying the difference between the error signal at the current moment and the error signal at the previous moment by the preset proportional coefficient, reflecting the magnitude of the error signal change between adjacent sampling periods. The proportional increment provides an immediate response to error changes; when the exposure deviation value changes abruptly, the proportional increment increases accordingly to accelerate the adjustment speed.

[0068] After the proportional increment calculation is completed, this embodiment calculates the integral increment based on a preset integral coefficient. The integral increment is obtained by multiplying the error signal at the current moment by the preset integral coefficient and is used to eliminate the steady-state deviation of the system. When the exposure deviation value persists, the integral increment continuously accumulates into the power adjustment amount, driving the infrared emission power to continuously adjust until the exposure deviation value approaches zero.

[0069] Based on the aforementioned proportional and integral increments, this embodiment calculates the differential increment according to a preset differential coefficient. The differential increment is obtained by subtracting twice the error signal from the previous moment and adding the error signals from two moments prior, then multiplying by the preset differential coefficient, reflecting the second-order characteristic of the error signal's changing trend. The differential increment provides a predictive response to the acceleration or deceleration trend of error changes; when the error change exhibits an oscillating tendency, the differential increment provides a damping effect to suppress overshoot. In this embodiment, the power adjustment amount is obtained by summing the proportional increment, integral increment, and differential increment.

[0070] After the power adjustment amount is calculated, this embodiment reads the current infrared emission power and adds it to obtain the target emission power. The current infrared emission power is the actual operating power of the infrared LED before this adjustment, and this embodiment reads this value from the status register of the infrared LED driver circuit. The target emission power represents the power level expected to be achieved after this adjustment, and is calculated by adding the current infrared emission power to the power adjustment amount.

[0071] Accordingly, this embodiment compares the target emission power with preset upper and lower power limits and performs constraint processing. The preset upper power limit defines the maximum allowable operating power of the infrared emitter; exceeding this value may damage the emitter or adversely affect facial tissue. The preset lower power limit defines the minimum effective operating power of the infrared emitter; below this value, the infrared light intensity will be insufficient to penetrate skin tissue to complete vascular imaging. This embodiment determines whether the target emission power exceeds the preset upper power limit; if so, it is constrained to the preset upper power limit. This embodiment further determines whether the target emission power is lower than the preset lower power limit; if so, it is constrained to the preset lower power limit. The emission power after constraint processing is within the allowable range, and this embodiment records it as the constrained emission power.

[0072] After the constrained emission power is determined, this embodiment sends it to the infrared LED driver circuit for power update. Upon receiving the constrained emission power value, the infrared LED driver circuit calculates the target drive current based on the mapping relationship between power and drive current. The driver circuit adjusts the duty cycle of the pulse width modulation signal to output the target drive current, and the emission power of the infrared LED is updated to the constrained emission power level as the drive current changes. After the power update is completed, subsequently acquired infrared images will reflect the adjusted exposure state, which can be read by subsequent step S501 for constructing multi-scale Gaussian smoothing spatial and vascular enhancement processing.

[0073] In one embodiment of the adaptive infrared blood vessel imaging enhancement method based on image entropy feedback in this application, it may further include the following: Step S501: Perform Gaussian convolution smoothing on the updated infrared image according to the preset scale parameter sequence to obtain a multi-scale Gaussian smoothing space. Calculate the horizontal second-order partial derivative, vertical second-order partial derivative, and mixed second-order partial derivative for each pixel in the multi-scale Gaussian smoothing space and assemble them into a Hessian matrix. Solve the characteristic equation of the Hessian matrix to obtain the first eigenvalue and the second eigenvalue. Step S502: Calculate the speckle discriminant factor and the structure sensitivity factor based on the first feature value and the second feature value. Substitute the speckle discriminant factor and the structure sensitivity factor into the tubular feature discriminant formula to calculate the vascular response value at each scale. Extract the maximum value of each pixel along the scale dimension from the vascular response values ​​at each scale to obtain the vascular enhancement image.

[0074] This embodiment reads the infrared image acquired after the power update in step S402 and performs Gaussian convolution smoothing on it sequentially according to a preset scale parameter sequence. The preset scale parameter sequence contains several scale factors arranged in ascending order, with smaller scale factors corresponding to the detection range of finer blood vessels and larger scale factors corresponding to the detection range of thicker blood vessels. In this embodiment, the starting and ending values ​​of the scale factors and the step interval are determined according to the target blood vessel diameter range, and the scale parameter sequence covers the complete blood vessel width range from small arteries to large veins.

[0075] After the preset scale parameter sequence is determined, this embodiment performs Gaussian convolution smoothing on the infrared image sequentially according to the sequence. Gaussian convolution smoothing constructs a two-dimensional Gaussian kernel function based on the current scale factor, and the spatial expansion range of the Gaussian kernel function is positively correlated with the scale factor. In this embodiment, the infrared image and the Gaussian kernel function are convolved, and the convolution result is a smoothed image at the current scale. The smoothed images corresponding to each scale factor are stacked according to the scale dimension to form a multi-scale Gaussian smoothing space. The multi-scale Gaussian smoothing space is stored in a three-dimensional data structure, with the first two dimensions corresponding to the image space coordinates and the third dimension corresponding to the scale parameter index.

[0076] Accordingly, this embodiment calculates the Hessian matrix for each pixel in the multi-scale Gaussian smoothed image at each scale. For the current pixel in the current scale image, this embodiment first calculates the second-order partial derivative in the horizontal direction, which is approximated by the gray-level difference of the pixel along the horizontal direction. This embodiment then calculates the second-order partial derivative in the vertical direction, which is approximated by the gray-level difference of the pixel along the vertical direction. This embodiment subsequently calculates the mixed second-order partial derivative, which is approximated by the gray-level difference of the pixel along the diagonal direction.

[0077] After the three second-order partial derivatives are calculated, this embodiment assembles them into a Hessian matrix. The Hessian matrix is ​​a second-order square matrix structure, with the horizontal and vertical second-order partial derivatives placed on the diagonal positions, and the mixed second-order partial derivatives placed on the off-diagonal positions. The Hessian matrix describes the local curvature features of the gray-level surface at the current pixel, and the tubular blood vessel structure exhibits a specific pattern in the eigenvalue distribution of the Hessian matrix.

[0078] Based on the aforementioned Hessian matrix, this embodiment solves its characteristic equation to obtain the first and second eigenvalues. The characteristic equation is solved by calculating the characteristic polynomial of the Hessian matrix and finding its roots; a second-order square matrix corresponds to two eigenvalue roots. In this embodiment, the two eigenvalues ​​are sorted by their absolute values, with the smaller absolute value recorded as the first eigenvalue and the larger absolute value recorded as the second eigenvalue. This sorting constraint ensures the consistency of subsequent discriminant factor calculations. The first and second eigenvalues ​​are written into the eigenvalue record of the current pixel for subsequent steps in S502.

[0079] In this embodiment, the first and second feature values ​​generated in step S501 are read, and a speckled discriminant factor is calculated based on them. The speckled discriminant factor is obtained by dividing the absolute value of the first feature value by the absolute value of the second feature value, and is used to distinguish between tubular structures and circular specks. For an ideal tubular vascular structure, the curvature along the direction of the vessel is close to zero, while the curvature perpendicular to the direction of the vessel is larger. The absolute value of the first feature value is much smaller than the absolute value of the second feature value, and the speckled discriminant factor approaches zero. For a circular speckled structure, the curvature in both directions is similar, and the speckled discriminant factor approaches one.

[0080] After the speckled discrimination factor is calculated, this embodiment calculates the structure sensitivity factor based on the first and second eigenvalues. The structure sensitivity factor is obtained by taking the square root of the sum of the squares of the first and second eigenvalues, reflecting the overall curvature intensity of the grayscale surface at the current pixel. The structure sensitivity factor is used to suppress the response of low-contrast background regions. Background regions exhibit a smaller structure sensitivity factor due to their gradual grayscale changes, while vascular regions exhibit a larger structure sensitivity factor due to their significant grayscale changes.

[0081] Accordingly, in this embodiment, the speckled discriminant factor and the structural sensitivity factor are substituted into the tubular feature discrimination formula to calculate the vascular response values ​​at each scale. The tubular feature discrimination formula can be expressed as: V = exp(-L L / (2 m1 m1)) (1 - exp(-S S / (2 m2 m2))).

[0082] In the formula, V is the vascular response value of the current pixel at the current scale; L is the speckle discrimination factor calculated above; S is the structure sensitivity factor calculated above; m1 and m2 are sensitivity control parameters, determined according to the vascular imaging characteristics during the system calibration stage. V achieves a higher value when the speckle discrimination factor approaches zero and the structure sensitivity factor is large, corresponding to a high-confidence response of tubular vascular structures.

[0083] After calculating the vascular response values ​​at each scale, this embodiment extracts the maximum value along the scale dimension for each pixel to obtain a vascular enhancement image. This embodiment iterates through all scale response values ​​corresponding to each pixel in a multi-scale Gaussian smoothed space, selecting the largest value as the final vascular response value for that pixel. This maximum value extraction operation ensures that the response value of each pixel corresponds to its optimal matching scale, allowing blood vessels of different thicknesses to obtain the strongest response at their respective suitable scales. This embodiment assembles the final vascular response values ​​of all pixels into a vascular enhancement image, which is then output to subsequent step S601 for adaptive threshold binarization and skeleton extraction processing.

[0084] In one embodiment of the adaptive infrared blood vessel imaging enhancement method based on image entropy feedback in this application, it may further include the following: Step S601: Divide the enhanced blood vessel image into local regions according to the preset neighborhood window size, statistically analyze the pixel grayscale distribution in each local region and calculate the local threshold, compare the grayscale value of each pixel with the corresponding local threshold, and mark the pixels with a value greater than the local threshold as the foreground and the pixels with a value less than the local threshold as the background to obtain a binary blood vessel mask. Step S602: Input the binary blood vessel mask into the iterative refinement algorithm. In each iteration, traverse the foreground pixels and determine whether the pixel is a deletable edge pixel based on the eight-neighbor connectivity condition. Set the edge pixels that meet the deletion condition as the background and retain the pixels that do not meet the deletion condition. Repeat the iteration until there are no deletable pixels to obtain a single-pixel width skeleton map.

[0085] This embodiment reads the enhanced blood vessel image generated in step S502 and divides it into local regions according to a preset neighborhood window size. The preset neighborhood window size defines the pixel span of the local region in the horizontal and vertical directions, and the size is determined based on the resolution and blood vessel distribution density of the enhanced blood vessel image. In this embodiment, the local region surrounding the current pixel is determined according to the preset neighborhood window size, and the local region covers a rectangular window formed by the current pixel and its neighboring pixels.

[0086] After the local regions are divided, this embodiment statistically analyzes the pixel grayscale distribution within each local region and calculates a local threshold. This embodiment iterates through all pixels within a local region and reads the grayscale value of each pixel. The grayscale values ​​are summed and divided by the number of pixels in the local region to obtain the local grayscale mean. This local grayscale mean serves as the basis for calculating the local threshold. This embodiment calculates the local threshold based on the local grayscale mean and a preset offset. Different local regions have different local thresholds due to differences in tissue background; high-brightness areas such as the forehead have higher local thresholds, while low-brightness areas such as the cheeks have lower local thresholds.

[0087] Accordingly, this embodiment compares the grayscale value of each pixel with the corresponding local threshold to complete the foreground and background labeling. For the current pixel, this embodiment reads its grayscale value and compares it with the local threshold of the local region to which the pixel belongs. If the pixel grayscale value is greater than the local threshold, this embodiment labels the pixel as foreground, and the foreground pixel corresponds to the vascular response region in the vascular enhancement image. If the pixel grayscale value is less than the local threshold, this embodiment labels the pixel as background, and the background pixel corresponds to the non-vascular region in the vascular enhancement image.

[0088] After the foreground and background markings are completed, this embodiment assembles the marking results of all pixels into a binary blood vessel mask. The binary blood vessel mask stores the category attributes of each pixel in binary form, with foreground pixels represented by the value 1 and background pixels represented by the value 0. The binary blood vessel mask preserves the morphological information of the blood vessel region in the enhanced blood vessel image, providing input data for subsequent skeleton extraction.

[0089] Based on the aforementioned binary blood vessel mask, this embodiment inputs it into an iterative thinning algorithm for skeleton extraction. The iterative thinning algorithm employs a layer-by-layer peeling strategy, progressively removing edge pixels while maintaining the topological connectivity of the blood vessels. In this embodiment, a deletion marker counter is initialized at the beginning of each iteration to record the number of edge pixels deleted in that iteration.

[0090] During a single iteration, this embodiment traverses all foreground pixels in the binary blood vessel mask and determines whether they are removable edge pixels based on the eight-neighbor connectivity condition. For the current foreground pixel, this embodiment reads the pixel values ​​of its eight neighboring locations and counts the number of foreground pixels within the neighborhood. The eight-neighbor connectivity condition includes two criteria: a neighborhood foreground pixel quantity constraint and a connectivity preservation constraint. The neighborhood foreground pixel quantity constraint requires that the number of foreground pixels within the neighborhood be within a preset range, while the connectivity preservation constraint requires that the foreground pixels within the neighborhood remain connected after the current pixel is deleted.

[0091] Accordingly, this embodiment performs deletion operations on edge pixels that meet the deletion conditions. If the current foreground pixel simultaneously satisfies both the neighboring foreground quantity constraint and the connectivity preservation constraint, this embodiment marks it as a deletable edge pixel and sets it to the background at the end of this iteration. If the current foreground pixel does not satisfy either constraint, this embodiment retains its foreground attributes without modification. After this round of iteration is completed, this embodiment counts the number of deleted edge pixels and updates the deletion mark counter.

[0092] After each iteration, this embodiment checks the value of the deletion marker counter to determine whether to continue iteration. If the deletion marker counter value is greater than zero, it indicates that there are deleted edge pixels in this iteration, and the blood vessel region has not yet shrunk to a single pixel width; this embodiment continues to execute the next iteration. If the deletion marker counter value is equal to zero, it indicates that there are no deleted edge pixels in this iteration, and the blood vessel region has shrunk to a single pixel width; this embodiment terminates the iteration process. The foreground pixels retained at the end of the iteration constitute a single-pixel width skeleton map, which is output to the subsequent step S701 for crossover calculation and blood vessel crossover point detection.

[0093] In one embodiment of the adaptive infrared blood vessel imaging enhancement method based on image entropy feedback in this application, it may further include the following: Step S701: Traverse each non-zero pixel in the single-pixel width skeleton diagram, read the pixel value sequence of the eight neighboring positions around the current pixel, calculate the absolute value of the difference between adjacent pixel values ​​in a clockwise direction, sum them up and divide by two to obtain the cross number, and store the cross number in association with the coordinates of the current pixel to form a pixel topology attribute record. Step S702: Filter pixel coordinates with a cross-number greater than a preset connection threshold from the pixel topology attribute record to obtain a set of blood vessel cross-point coordinates. Generate highlight mark graphic elements according to the coordinate positions of each blood vessel cross-point coordinate set and assemble them into a highlight mark layer. Align the highlight mark layer with the current imaging screen and overlay it to the display terminal for output.

[0094] This embodiment reads the single-pixel width skeleton map generated in step S602 and iterates through each non-zero pixel to perform cross-multiplication calculation. The single-pixel width skeleton map is stored in binary form, with non-zero pixels corresponding to the location of the blood vessel skeleton and zero-value pixels corresponding to the background area. This embodiment scans all pixel positions of the single-pixel width skeleton map in row and column order, and initiates the cross-multiplication calculation process for positions with non-zero pixel values.

[0095] After locating the non-zero pixel, this embodiment reads the pixel value sequence of the eight neighboring positions surrounding the current pixel. The eight neighboring positions include the positions above, below, to the left, to the right, and in the four diagonal directions of the current pixel. In this embodiment, the pixel values ​​of each neighboring position are read sequentially, starting from the right position in a clockwise direction. The pixel value sequence contains eight elements, each with a value of either zero or one, reflecting whether the corresponding neighboring position belongs to the vascular skeleton.

[0096] Accordingly, this embodiment calculates the absolute value of the difference between adjacent pixel values ​​sequentially in a clockwise direction. This embodiment treats the pixel value sequence as a circular structure with its ends connected, calculating the absolute value of the difference between the first element and the next, starting from the first element, until the absolute value of the difference between the last element and the first element is calculated. The calculation process generates eight absolute difference values, each reflecting the change in the skeleton connectivity between adjacent neighborhood positions.

[0097] After the crossover count is calculated, this embodiment associates it with the current pixel coordinates to form a pixel topology attribute record. The pixel topology attribute record is organized in a data structure, containing the horizontal and vertical coordinates of the pixel and the corresponding crossover count value. This embodiment writes the topology attribute record of the current non-zero pixel into a pixel topology attribute record set. After traversal, the set contains the topology attribute information of all non-zero pixels in the single-pixel width skeleton graph.

[0098] Based on the aforementioned pixel topology attribute record set, this embodiment filters pixel coordinates with a crossover number greater than a preset connection threshold. The preset connection threshold is used to distinguish between ordinary connection points and crossover / bifaction points. In this embodiment, the preset connection threshold is set so that pixels with a crossover number greater than this threshold are determined to be either crossover points or bifaction points. This embodiment traverses each record in the pixel topology attribute record set, reads its crossover number value and compares it with the preset connection threshold, and extracts the pixel coordinates from records with a crossover number greater than the preset connection threshold into the blood vessel crossover point coordinate set.

[0099] After the set of coordinates of the blood vessel intersections is generated, this embodiment generates highlighted graphic elements based on each coordinate position. The highlighted graphic elements use a preset geometric configuration, and this embodiment creates an independent graphic instance for each intersection coordinate position. The color attribute of the graphic is set to a high-contrast color for easy identification by the operator, and the size attribute of the graphic is determined according to the display resolution and viewing distance. This embodiment assembles the graphic elements corresponding to all intersections into a highlighted layer.

[0100] After the highlighted marker layer is assembled, this embodiment aligns it with the current imaging screen using coordinates. The coordinate alignment process matches spatial positions based on the mapping relationship between the image coordinate system and the display coordinate system. This embodiment calculates the corresponding position of each marker graphic in the highlighted marker layer in the display coordinate system and updates its coordinate attributes. After coordinate alignment, the display position of each marker graphic in the highlighted marker layer coincides with the position of the corresponding blood vessel intersection point in the imaging screen. This embodiment overlays and renders the aligned highlighted marker layer onto the display terminal for output. Operators can visually observe the distribution of intersection points in the facial blood vessel network through the display terminal, avoiding these intersection points when performing facial area operations to reduce operational risks.

[0101] To effectively address the shortcomings of traditional technologies in areas such as exposure adaptation, vessel enhancement, and topology resolution, and to provide technical support for infrared vascular imaging, this application provides an embodiment of an adaptive infrared vascular imaging enhancement device based on image entropy feedback for implementing all or part of the aforementioned adaptive infrared vascular imaging enhancement method based on image entropy feedback. See [link to embodiment]. Figure 2 The adaptive infrared vascular imaging enhancement device based on image entropy feedback specifically includes the following components: The pixel calculation module 10 is used to acquire the original infrared image of the face region through the infrared image acquisition device and divide it into several sub-regions according to the preset grid size. It calculates the pixel gray-scale mean and gray-scale probability distribution for each sub-region and substitutes them into the information entropy formula to obtain the local information entropy. It weights and sums the pixel gray-scale mean and the local information entropy according to the preset weight coefficient to obtain the exposure score. It compares the exposure score with the preset ideal imaging range to calculate the exposure deviation value. Feature processing module 20 is used to input the exposure deviation value as an error signal into an incremental proportional-integral-derivative controller to calculate the power adjustment amount, superimpose the power adjustment amount onto the current infrared emission power and send it to the infrared emitting tube drive circuit to perform power update, construct a multi-scale Gaussian smooth space for the updated infrared image and calculate the Hessian matrix and eigenvalues ​​at each scale, substitute the eigenvalues ​​into the tubular feature discrimination formula to calculate the vascular response value at each scale and take the maximum value in the scale dimension to obtain the vascular enhancement image; Infrared imaging module 30 is used to perform adaptive threshold binarization on the enhanced blood vessel image to obtain a binary blood vessel mask, input the binary blood vessel mask into an iterative thinning algorithm to strip edge pixels to obtain a single-pixel width skeleton map, traverse the eight-neighborhood states of each non-zero pixel in the single-pixel width skeleton map and calculate the number of intersections, extract the coordinates of pixels with an intersection number greater than a preset connection threshold as a set of blood vessel intersection coordinates and render them as a highlight mark layer superimposed on the display terminal for output.

[0102] As described above, the adaptive infrared vascular imaging enhancement device based on image entropy feedback provided in this application can achieve precise adaptive adjustment of infrared emission power through local information entropy calculation and a proportional-integral-differential controller. An enhancement mechanism is constructed, combining multi-scale Gaussian smoothing and Hessian matrix eigenvalue analysis to establish a reliable vascular response extraction strategy. Structural analysis optimization is introduced, and continuous improvement in vascular intersection point localization is ensured through iterative refinement of skeleton extraction and intersection number calculation. This method effectively addresses the shortcomings of traditional techniques in exposure adaptation, vascular enhancement, and topology analysis, providing technical support for infrared vascular imaging-assisted diagnosis.

[0103] This invention also provides a computer device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the adaptive infrared vascular imaging enhancement method based on image entropy feedback.

[0104] This invention also provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the above-described adaptive infrared vascular imaging enhancement method based on image entropy feedback.

[0105] This invention also provides a computer program product, which includes a computer program that, when executed by a processor, implements the above-described adaptive infrared vascular imaging enhancement method based on image entropy feedback.

[0106] In this embodiment of the invention, precise adaptive adjustment of infrared emission power is achieved through local information entropy calculation and a proportional-integral-differential controller. An enhancement mechanism is constructed, combining multi-scale Gaussian smoothing and Hessian matrix eigenvalue analysis to establish a reliable vascular response extraction strategy. Structural analytical optimization is introduced, and continuous improvement in vascular intersection point localization is ensured through iterative refinement of skeleton extraction and intersection number calculation. This method effectively addresses the shortcomings of traditional techniques in exposure adaptation, vascular enhancement, and topology analysis, providing technical support for infrared vascular imaging-assisted diagnosis.

[0107] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0108] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0109] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0110] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0111] The specific embodiments described above further illustrate the purpose, technical solution, and beneficial effects of the present invention. It should be understood that the above descriptions are merely specific embodiments of the present invention and are not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. An adaptive infrared vascular imaging enhancement method based on image entropy feedback, characterized in that, The method includes: The original infrared image of the face region is acquired by an infrared image acquisition device and divided into several sub-regions according to a preset grid size. The mean gray value and gray level probability distribution of each sub-region are calculated and substituted into the information entropy formula to obtain the local information entropy. The mean gray value of the pixels and the local information entropy are weighted and summed according to a preset weight coefficient to obtain the exposure score. The exposure score is compared with the preset ideal imaging range to calculate the exposure deviation value. The exposure deviation value is used as an error signal input to the incremental proportional-integral-derivative controller to calculate the power adjustment amount. The power adjustment amount is superimposed on the current infrared emission power and sent to the infrared emission tube drive circuit to perform power update. A multi-scale Gaussian smooth space is constructed for the updated infrared image and the Hessian matrix and eigenvalues ​​are calculated at each scale. The eigenvalues ​​are substituted into the tubular feature discrimination formula to calculate the blood vessel response value at each scale and the maximum value is taken in the scale dimension to obtain the blood vessel enhancement image. Adaptive threshold binarization is performed on the enhanced blood vessel image to obtain a binary blood vessel mask. The binary blood vessel mask is input into an iterative thinning algorithm to strip edge pixels and obtain a single-pixel width skeleton map. The eight-neighborhood states of each non-zero pixel in the single-pixel width skeleton map are traversed and the number of intersections is calculated. The coordinates of pixels with an intersection number greater than a preset connection threshold are extracted as a set of blood vessel intersection coordinates and rendered as a highlight mark layer, which is then superimposed on the display terminal for output.

2. The adaptive infrared vascular imaging enhancement method based on image entropy feedback according to claim 1, characterized in that, The process involves acquiring the original infrared image of the facial region using an infrared image acquisition device, dividing it into several sub-regions according to a preset grid size, calculating the mean pixel grayscale value and grayscale probability distribution for each sub-region, and substituting these into the information entropy formula to obtain the local information entropy, including: The face area is scanned and acquired by an infrared image acquisition device to obtain an original infrared image. The original infrared image is evenly divided into several sub-regions in the horizontal and vertical directions according to a preset grid size. The pixel gray values ​​in each sub-region are summed and divided by the total number of pixels in the sub-region to obtain the pixel gray average. The number of pixels at each gray level in each sub-region is counted and divided by the total number of pixels in the sub-region to obtain the gray level probability distribution. Read the probability value corresponding to each gray level in the gray level probability distribution, take the logarithm of each probability value and multiply it by the probability value to obtain the entropy contribution of each gray level, sum the entropy contributions of each gray level and take the negative value to obtain the local information entropy, and associate the pixel gray mean with the local information entropy to form sub-region feature data for subsequent exposure score calculation.

3. The adaptive infrared vascular imaging enhancement method based on image entropy feedback according to claim 1, characterized in that, The step of obtaining an exposure score by weighting and summing the average pixel grayscale value and the local information entropy according to a preset weighting coefficient, and calculating the exposure deviation value by comparing the exposure score with a preset ideal imaging range, includes: The average pixel grayscale value and local information entropy are read from the sub-region feature data. The average pixel grayscale value and the local information entropy are weighted according to the preset weight coefficient and summed to obtain the exposure score. The preset weight coefficient consists of brightness weight and feature richness weight, and the sum of the two is a unit value. The target grayscale expectation value is read from the preset ideal imaging range. The exposure score is subtracted from the target grayscale expectation value to obtain the exposure deviation value. The exposure deviation value is output to the power adjustment stage as an error signal for the incremental proportional-integral-derivative controller to call.

4. The adaptive infrared vascular imaging enhancement method based on image entropy feedback according to claim 1, characterized in that, The step of using the exposure deviation value as an error signal to input an incremental proportional-integral-derivative controller to calculate a power adjustment amount, adding the power adjustment amount to the current infrared emission power, and sending it to the infrared emitting tube drive circuit to perform a power update includes: The exposure deviation value is used as the error signal at the current moment and input to the incremental proportional-integral-derivative controller. The error signal at the current moment, the error signal at the previous moment, and the error signal at the two moments before that are read. The proportional increment, integral increment, and derivative increment are calculated according to the preset proportional coefficient, integral coefficient, and derivative coefficient, and then summed to obtain the power adjustment amount. The current infrared emission power is read and the power adjustment amount is added to the current infrared emission power to obtain the target emission power. The target emission power is compared with the preset power upper limit value and the preset power lower limit value and constrained to the allowable range to obtain the constrained emission power. The constrained emission power is sent to the infrared emitting tube drive circuit to perform power update.

5. The adaptive infrared vascular imaging enhancement method based on image entropy feedback according to claim 1, characterized in that, The process involves constructing a multi-scale Gaussian smoothed space from the updated acquired infrared images, calculating the Hessian matrix and eigenvalues ​​at each scale, substituting the eigenvalues ​​into the tubular feature discrimination formula to calculate the vascular response values ​​at each scale, and taking the maximum value in the scale dimension to obtain the vascular enhancement image. This includes: The updated infrared images are sequentially smoothed by Gaussian convolution according to a preset scale parameter sequence to obtain a multi-scale Gaussian smooth space. In the images at each scale of the multi-scale Gaussian smooth space, the second-order partial derivatives in the horizontal direction, the second-order partial derivatives in the vertical direction, and the mixed second-order partial derivatives are calculated for each pixel and assembled into a Hessian matrix. The characteristic equation of the Hessian matrix is ​​solved to obtain the first eigenvalue and the second eigenvalue. Based on the first feature value and the second feature value, a speckled discriminant factor and a structural sensitivity factor are calculated. The speckled discriminant factor and the structural sensitivity factor are substituted into the tubular feature discriminant formula to calculate the vascular response value at each scale. The maximum value of each pixel is extracted along the scale dimension from the vascular response values ​​at each scale to obtain the vascular enhancement image.

6. The adaptive infrared vascular imaging enhancement method based on image entropy feedback according to claim 1, characterized in that, The process of performing adaptive threshold binarization on the enhanced blood vessel image to obtain a binary blood vessel mask, and inputting the binary blood vessel mask into an iterative thinning algorithm to remove edge pixels to obtain a single-pixel width skeleton map, includes: The enhanced vascular image is divided into local regions according to a preset neighborhood window size. The pixel grayscale distribution is statistically analyzed in each local region and a local threshold is calculated. The grayscale value of each pixel is compared with the corresponding local threshold, and pixels with a value greater than the local threshold are marked as foreground and pixels with a value less than the local threshold are marked as background to obtain a binary vascular mask. The binary blood vessel mask is input into an iterative refinement algorithm. In each iteration, the foreground pixels are traversed and it is determined whether the pixel is a deletable edge pixel based on the eight-neighbor connectivity condition. Edge pixels that meet the deletion condition are set as background and pixels that do not meet the deletion condition are retained. The iteration is repeated until there are no deletable pixels to obtain a single-pixel width skeleton map.

7. The adaptive infrared vascular imaging enhancement method based on image entropy feedback according to claim 1, characterized in that, The process of traversing the eight-neighbor states of each non-zero pixel in the single-pixel width skeleton graph and calculating the number of intersections, extracting the pixel coordinates of those with an intersection number greater than a preset connection threshold as a set of blood vessel intersection coordinates, rendering them as a highlight mark layer, and overlaying them onto the display terminal for output includes: Traverse each non-zero pixel in the single-pixel width skeleton diagram, read the pixel value sequence of the eight neighboring positions around the current pixel, calculate the absolute value of the difference between adjacent pixel values ​​in a clockwise direction, sum them and divide by two to obtain the cross number, and store the cross number in association with the coordinates of the current pixel to form a pixel topology attribute record. The set of blood vessel intersection coordinates is obtained by filtering pixel coordinates with a number of intersections greater than a preset connection threshold from the pixel topology attribute record. Highlighted marker graphic elements are generated according to the coordinate positions of each coordinate in the set of blood vessel intersection coordinates and assembled into a highlighted marker layer. The highlighted marker layer is aligned with the current imaging screen and superimposed and rendered to the display terminal for output.

8. An adaptive infrared vascular imaging enhancement device based on image entropy feedback, characterized in that, The device includes: The pixel calculation module is used to acquire the original infrared image of the face area through the infrared image acquisition device and divide it into several sub-regions according to the preset grid size. It calculates the pixel gray mean and gray level probability distribution for each sub-region and substitutes them into the information entropy formula to obtain the local information entropy. It weights and sums the pixel gray mean and the local information entropy according to the preset weight coefficient to obtain the exposure score. It compares the exposure score with the preset ideal imaging range to calculate the exposure deviation value. The feature processing module is used to input the exposure deviation value as an error signal into the incremental proportional-integral-derivative controller to calculate the power adjustment amount, superimpose the power adjustment amount onto the current infrared emission power and send it to the infrared emitting tube drive circuit to perform power update, construct a multi-scale Gaussian smooth space for the updated infrared image and calculate the Hessian matrix and eigenvalues ​​at each scale, substitute the eigenvalues ​​into the tubular feature discrimination formula to calculate the blood vessel response value at each scale and take the maximum value in the scale dimension to obtain the blood vessel enhancement image; The infrared imaging module is used to perform adaptive threshold binarization on the enhanced blood vessel image to obtain a binary blood vessel mask. The binary blood vessel mask is input into an iterative thinning algorithm to strip edge pixels to obtain a single-pixel width skeleton map. The eight-neighborhood states of each non-zero pixel in the single-pixel width skeleton map are traversed and the number of intersections is calculated. The coordinates of pixels with an intersection number greater than a preset connection threshold are extracted as a set of blood vessel intersection coordinates and rendered as a highlight mark layer, which is then superimposed on the display terminal for output.

9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the program, it implements the steps of the adaptive infrared vascular imaging enhancement method based on image entropy feedback as described in any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When executed by a processor, the computer program implements the steps of the adaptive infrared vascular imaging enhancement method based on image entropy feedback as described in any one of claims 1 to 7.

Images