Computer image processing system based on deep learning

Abstract

The invention relates to the technical field of image processing, in particular to a computer image processing system based on deep learning, which determines a first contour boundary of a target object through a preprocessing module according to a deep learning model, and draws brightness expression curves of pixel points of each sub-region in different direction dimensions through a feature extraction module. Marking a non-confidence sub-region in the aggregation analysis region through a feature aggregation module, obtaining brightness expression curves of pixel points subjected to exposure adjustment in different direction dimensions through a local processing module so as to determine contour boundary fitting points, and fitting contour boundary sections of the non-confidence sub-region, and updating the first contour boundary through the output end and outputting a second contour boundary of the target object. According to the method, local region optimization is carried out aiming at insufficient local features in a target object contour boundary extraction process under a complex illumination condition, and high-precision extraction and dynamic updating of a target object contour in a complex environment are realized.

Description

Technical Field

[0001] The present invention relates to the field of image processing technology, and in particular to a computer image processing system based on deep learning. Background Art

[0002] In the field of computer image processing, accurate extraction of target contour boundaries is the basis for tasks such as image analysis, object recognition, and scene understanding. Traditional image processing methods often rely on fixed thresholds, edge detection operators, etc. to achieve contour extraction. However, such methods are prone to problems such as contour breakage, false detection, or detail loss in scenes with low contrast between the target and the background, blurred texture features, or noise interference. In complex situations such as areas with sudden brightness changes under non-uniform illumination and similar features between the target edge and the background, contour extraction models based on single global features still find it difficult to accurately capture boundary details. In existing technologies, deep learning models lack detailed analysis of local area brightness features, especially when dealing with the problem of blurred contour boundaries caused by exposure differences. There is a lack of targeted optimization mechanisms for untrusted areas. Therefore, there is an urgent need for an image processing system that combines deep learning models with local brightness feature analysis to solve the technical problem of insufficient accuracy in target contour boundary extraction under complex lighting conditions.

[0003] For example, Chinese patent publication number: CN119338745A, the invention discloses a computer image processing system based on intelligent recognition technology, including: a real-time monitoring module, which uses real-time image quality assessment technology to monitor and learn the acquired image data information; an alarm system is set in the image recognition module, and the alarm system is triggered according to the change of the acquired image data information; the image optimization module receives the alarm signal, locks the damaged position of the image, intercepts the image information before and after the damage of the image, and uses a degraded image restoration algorithm based on a neural network to deeply learn the image information; uses a degraded image restoration algorithm based on a neural network to deeply learn the image information, extract clear original information, and perform blurring and denoising processing to effectively fill the damaged position of the image.

[0004] The following problems also exist in the prior art:

[0005] Existing technologies cannot optimize local areas to address the problem of insufficient local features during target object contour boundary extraction under complex lighting conditions, which affects the accuracy of contour extraction. Summary of the Invention

[0006] To this end, the present invention provides a computer image processing system based on deep learning to overcome the problem that the existing technology cannot optimize local areas based on insufficient local features during the process of extracting the contour boundaries of target objects under complex lighting conditions.

[0007] To achieve the above objectives, the present invention provides a computer image processing system based on deep learning, comprising:

[0008] a preprocessing module configured to divide the reference image into a plurality of rectangular sub-regions, determine a first contour boundary of the target object based on a deep learning model, and mark position information of the sub-regions where contour boundary segments exist;

[0009] A feature extraction module, which is connected to the preprocessing module and is used to obtain the brightness of the pixels in each sub-region to draw brightness performance curves of the pixels in each sub-region in different directions and dimensions;

[0010] a feature aggregation module, connected to the preprocessing module and the feature extraction module, respectively, for determining an aggregate analysis area based on position information matching conditions, and marking untrusted sub-areas according to the comparison of brightness performance curves of each sub-area within the aggregate analysis area;

[0011] a local processing module connected to the feature aggregation module, configured to obtain brightness performance curves of the pixels of the non-confident sub-region in different directions after the exposure is adjusted, and determine contour boundary fitting points based on comparison of the brightness performance curves after the exposure is adjusted, so as to fit the contour boundary segments of the non-confident sub-region;

[0012] An output end is connected to the local processing module and the pre-processing module respectively, and is used to update the first contour boundary using the contour boundary segment of the non-confidence sub-region to output a second contour boundary of the target object in the reference image.

[0013] Furthermore, the feature extraction module is used to calculate the average brightness of the pixels of each sub-region in different directions, wherein:

[0014] The feature extraction module is used to use the width data of each sub-region as the horizontal axis of the coordinate system to calculate the average brightness of the pixel points determined based on each width value point in the sub-region;

[0015] And the length data of each sub-region is used as the horizontal axis of the coordinate system to calculate the average brightness of the pixel points determined based on each length value point in the sub-region.

[0016] Furthermore, the feature extraction module is used to draw brightness performance curves in different directional dimensions, wherein:

[0017] The feature extraction module is used to draw a first brightness performance curve in a first direction based on the average brightness of the pixel points determined by the width value points, and to draw a second brightness performance curve in a second direction based on the average brightness of the pixel points determined by the length value points;

[0018] The horizontal axis of the first brightness performance curve is the width value of the sub-region, and the vertical axis is the average brightness of the pixels. The horizontal axis of the second brightness performance curve is the length value of the sub-region, and the vertical axis is the average brightness of the pixels.

[0019] Furthermore, the feature aggregation module is used to determine the aggregation analysis area based on the location information matching condition, wherein:

[0020] The feature aggregation module is used to obtain position information of each sub-region where a contour boundary segment exists, including a row sequence determined along the length direction of the reference image and a column sequence determined along the width direction of the reference image according to the position information;

[0021] The feature aggregation module is used to determine an area composed of several sub-areas that meet the location information matching condition as the aggregation analysis area;

[0022] The aggregate analysis area includes at least two sub-areas, and the position information matching condition is that the row sequence or column sequence of each sub-area is consistent.

[0023] Furthermore, the feature aggregation module is used to determine the comparison of brightness performance curves of each sub-region, wherein:

[0024] The feature aggregation module is used to determine a maximum brightness difference based on a brightness performance curve of each sub-region within the aggregate analysis region, so as to calculate a standard deviation of the maximum brightness difference of the sub-regions within the aggregate analysis region.

[0025] Furthermore, the feature aggregation module is used to mark untrusted sub-regions, wherein:

[0026] The feature aggregation module is used to screen the aggregation analysis area corresponding to the maximum value of the standard deviation, and mark the sub-areas corresponding to the maximum value and the minimum value of the maximum brightness difference in the corresponding aggregation analysis area as untrusted sub-areas.

[0027] Furthermore, the local processing module is used to obtain the brightness performance curves of the pixels in the non-confidence sub-region after adjusting the exposure in different directions, wherein:

[0028] The local processing module is configured to draw a third brightness performance curve in the first direction based on the average brightness of the pixels determined at each width value point in the non-confident sub-region after the exposure adjustment, and to draw a fourth brightness performance curve in the second direction based on the average brightness of the pixels determined at each length value point in the non-confident sub-region after the exposure adjustment;

[0029] The exposure of the non-confidence sub-region after adjustment is higher than the exposure before adjustment.

[0030] Furthermore, the local processing module is used to determine the contour boundary fitting points, wherein,

[0031] The local processing module is configured to compare the third brightness performance curve with the brightness average value of the first brightness performance curve on the vertical axis to determine a width value of a maximum difference in the brightness average value on the horizontal axis, and to compare the fourth brightness performance curve with the brightness average value of the second brightness performance curve on the vertical axis to determine a length value of the maximum difference in the brightness average value on the horizontal axis;

[0032] The local processing module is configured to determine the contour boundary fitting point in the non-confidence sub-region based on the width value and the length value.

[0033] Furthermore, the local processing module is used to fit the contour boundary segment of the non-confident sub-region, wherein,

[0034] The local processing module is used to obtain the initial contour boundary segment of the non-confidence sub-region, and fit the initial contour boundary segment based on the contour boundary fitting point, so that the fitted contour boundary segment passes through the contour boundary fitting point.

[0035] Furthermore, the output end is used to update the first contour boundary using the contour boundary segment that has completed fitting, wherein,

[0036] The output end is used to replace the initial contour boundary segment with the contour boundary segment that has completed fitting, so as to update the first contour boundary to the second contour boundary.

[0037] Compared with the prior art, the beneficial effect of the present invention lies in that the present invention determines the first contour boundary of the target object according to the deep learning model through the preprocessing module, and marks the position information of the sub-region where the contour boundary segment exists, draws the brightness performance curve of the pixel points of each sub-region in different directional dimensions through the feature extraction module, marks the non-confidence sub-region in the aggregation analysis area through the feature aggregation module, obtains the brightness performance curve of the pixel points of the non-confidence sub-region after adjusting the exposure in different directional dimensions through the local processing module, fits the contour boundary segment of the non-confidence sub-region according to the contour boundary fitting point, updates the first contour boundary through the output end and outputs the second contour boundary of the target object. The present invention optimizes the local area to address the lack of local features in the process of extracting the contour boundary of the target object under complex lighting conditions, and realizes high-precision extraction and dynamic update of the target object contour under complex lighting conditions.

[0038] Furthermore, the present invention divides the reference image into several rectangular sub-regions and performs brightness statistics along both the width and length of each sub-region. In the width direction, the average brightness of each column of pixels is calculated; in the length direction, the average brightness of each row of pixels is calculated. This two-dimensional analysis captures the brightness variation patterns in different directions within the sub-region. A first brightness curve is plotted to reflect the brightness distribution characteristics in the width direction, while a second brightness curve is plotted to reflect the brightness distribution characteristics in the length direction.

[0039] Furthermore, based on the phenomenon that the contour boundary of a target object is usually composed of adjacent or continuously arranged sub-regions, the present invention defines the aggregate analysis area according to the matching conditions of row consistency or column consistency through the feature aggregation module: if multiple sub-regions have the same row sequence or column sequence, they are divided into one aggregate analysis area. This strategy of aggregating and analyzing sub-regions in the same row or column can effectively capture the linear extension characteristics of the contour boundary and improve the continuity of the contour boundary analysis.

[0040] Furthermore, the present invention calculates the maximum brightness difference in the corresponding dimension for the brightness performance curve of each sub-region in the aggregation analysis area through the feature aggregation module. The difference reflects the severity of the brightness change in the sub-region. The standard deviation of the maximum brightness difference of all sub-regions in the aggregation area is calculated. The larger the standard deviation, the more significant the difference in brightness change of each sub-region in the aggregation analysis area, that is, the brightness characteristics of some sub-regions in the contour boundary of the same row or column are inconsistent with the overall trend. The aggregation analysis area with the largest standard deviation among all aggregation areas is screened out. The brightness feature consistency of the sub-regions in the aggregation analysis area is the worst, and may include contour blurring due to complex lighting, resulting in abnormal identified contour boundary segments. The feature consistency is quantified by standard deviation and combined with the extreme value positioning of abnormal sub-regions. The screening method can effectively distinguish reliable boundary segments from unreliable boundary segments in complex lighting scenes, thereby improving the robustness of overall contour extraction.

[0041] Furthermore, the present invention increases the exposure of the marked non-trusted sub-region, thereby amplifying the pixel brightness differences in the region that were originally masked by underexposure, especially the brightness gradient change between the target object and the background will be more obvious. After the exposure is enhanced, the module recalculates the average pixel brightness of each coordinate point in the width and length directions of the sub-region, and draws the third brightness performance curve and the fourth brightness performance curve respectively. Since the increase in exposure will cause the pixel values ​​of the dark area to shift to the high brightness range, the brightness mutation characteristics at the originally blurred boundary will become more obvious, thereby achieving the highlighting of boundary characteristics through local exposure enhancement.

[0042] Furthermore, the present invention plots a first and second brightness performance curve for a non-confident sub-region before exposure adjustment, and then generates third and fourth brightness performance curves after exposure adjustment. A local processing module compares the curves before and after adjustment on the vertical axis, identifying the location with the largest difference between the two. In the width direction, the horizontal axis width corresponding to the maximum difference between the average brightness of the third and first curves reflects the location with the most significant brightness change after exposure enhancement in that direction, typically corresponding to the boundary between the target and the background. Similarly, the length direction length corresponds to the maximum difference between the fourth and second curves. The point determined by the intersection of the determined width and length values ​​in the sub-region coordinate system is the contour boundary fitting point. Those skilled in the art will understand that exposure adjustment amplifies the brightness difference between the target and the background, and this difference is often more dramatic at the boundary than in non-boundary areas. By capturing the extreme difference locations in the curve comparison, boundary points that were originally blurred due to lighting issues can be accurately located. This invention utilizes the complementary information before and after exposure adjustment, avoids the limitations of contour feature extraction, and achieves local region optimization for insufficient local features. BRIEF DESCRIPTION OF THE DRAWINGS

[0043] Figure 1 This is a block diagram of a computer image processing system based on deep learning according to an embodiment of the present invention;

[0044] Figure 2 This is a schematic diagram of the sub-region distribution according to an embodiment of the present invention;

[0045] Figure 3 This is a logic flow chart of determining the aggregation analysis area by the feature aggregation module according to an embodiment of the present invention. DETAILED DESCRIPTION

[0046] In order to make the objects and advantages of the present invention more clearly understood, the present invention is further described below in conjunction with embodiments; it should be understood that the specific embodiments described herein are merely used to explain the present invention and are not intended to limit the present invention.

[0047] The preferred embodiments of the present invention are described below with reference to the accompanying drawings. It should be understood by those skilled in the art that these embodiments are only used to explain the technical principles of the present invention and are not intended to limit the scope of protection of the present invention.

[0048] It should be noted that, in the description of the present invention, terms such as "upper", "lower", "inside" and "outside" indicating directions or positional relationships are based on the directions or positional relationships shown in the accompanying drawings. This is only for the convenience of description and does not indicate or imply that the device or element must have a specific orientation, be constructed and operated in a specific orientation. Therefore, it cannot be understood as a limitation on the present invention.

[0049] Furthermore, it should be noted that, in the description of the present invention, unless otherwise expressly specified or limited, the terms "mounted" and "connected" should be understood in a broad sense. For example, they may refer to fixed, detachable, or integral connections; mechanical or electrical connections; direct or indirect connections through an intermediate medium; or internal communication between two components. Those skilled in the art will understand the specific meanings of the above terms in the present invention based on specific circumstances.

[0050] See also Figure 1 As shown in FIG, which is a block diagram of a computer image processing system based on deep learning according to an embodiment of the present invention, the computer image processing system based on deep learning according to the present invention includes:

[0051] a preprocessing module configured to divide the reference image into a plurality of rectangular sub-regions, determine a first contour boundary of the target object based on a deep learning model, and mark position information of the sub-regions where contour boundary segments exist;

[0052] The reference image in the embodiment of the present invention is an image that has been cropped and denoised. Cropping and denoising are common operations in the image processing process and will not be described in detail here.

[0053] Specifically, the preprocessing module in the implementation of the present invention can be a processor that integrates image segmentation and contour extraction functions, which is used to divide the reference image into several rectangular sub-areas of preset sizes using a grid division method. Preferably, the length value of the rectangular sub-area is 1 / 20 of the length value of the reference image, and the width value of the rectangular sub-area is 1 / 20 of the width value of the reference image.

[0054] Specifically, the preprocessing module determines the row sequence of each sub-region along the length direction of the reference image, determines the column sequence of each sub-region along the width direction of the reference image, and determines the position information of the sub-region where the contour boundary segment exists based on the row sequence and the column sequence.

[0055] Specifically, the deep learning model in the preprocessing module is pre-trained and is used to identify the first contour boundary of the target object. Those skilled in the art can understand that the training process of the deep learning model can be: pre-acquire multiple training samples, each of the multiple training samples includes information about an image and a label corresponding to the training sample, and the label corresponding to the training sample refers to the first contour boundary of the target object in the image. The model processor trains the initialized learning model based on the multiple training samples and the labels corresponding to each training sample to obtain a deep learning model. The deep learning model may include convolution layers, nonlinear activation layers, normalization layers, and deconvolution layers, etc. The construction layer of the deep learning model is not limited here. The construction layer of the deep learning model in the implementation of the present invention is constructed based on the above method, and the construction layer of the deep learning model is not explained here.

[0056] Preferably, each training sample includes each pixel of a reference image under different lighting conditions. The reference image information includes an RGB image or a grayscale image. In the case of an RGB image, the pixel values ​​of the image may be the pixel values ​​corresponding to the R channel, G channel, and B channel of each pixel in the RGB image. In the case of a grayscale image, the pixel values ​​of the image are the grayscale values ​​of each pixel in the grayscale image. The training process of the deep learning model is to train the initialized learning model based on the results of the recognition of each pixel and pixel value of the collected reference image and the first contour result of the target object in the image to obtain a deep learning model.

[0057] A feature extraction module, which is connected to the preprocessing module and is used to obtain the brightness of the pixels in each sub-region to draw brightness performance curves of the pixels in each sub-region in different directions and dimensions;

[0058] Specifically, the feature extraction module in the present invention can be a processor that integrates pixel brightness analysis and curve construction functions. In the implementation of the present invention, the brightness of the pixel point can be converted from RGB to grayscale values. This is a prior art and will not be repeated here.

[0059] a feature aggregation module, connected to the preprocessing module and the feature extraction module, respectively, for determining an aggregate analysis area based on position information matching conditions, and marking untrusted sub-areas according to the comparison of brightness performance curves of each sub-area within the aggregate analysis area;

[0060] Specifically, the feature aggregation module in the present invention may be an image processor for performing region marking processing on an image. This is a prior art and will not be described in detail here.

[0061] a local processing module connected to the feature aggregation module, configured to obtain brightness performance curves of the pixels of the non-confident sub-region in different directions after the exposure is adjusted, and determine contour boundary fitting points based on comparison of the brightness performance curves after the exposure is adjusted, so as to fit the contour boundary segments of the non-confident sub-region;

[0062] Specifically, the local processing module in the present invention is a processor with a curve fitting function. It determines the discrete points on the contour based on the known contour, and re-fits the contour boundary in combination with the newly determined contour boundary fitting points. This is achievable by the existing image processing algorithm and will not be repeated here.

[0063] An output end is connected to the local processing module and the pre-processing module respectively, and is used to update the first contour boundary using the contour boundary segment of the non-confidence sub-region to output a second contour boundary of the target object in the reference image.

[0064] Specifically, the output end in the present invention may be a display, which displays and outputs the processed image contour by receiving the signal after image contour processing, and will not be described in detail here.

[0065] Specifically, the feature extraction module is used to calculate the average brightness of the pixels in each sub-region in different directions, where:

[0066] The feature extraction module is used to use the width data of each sub-region as the horizontal axis of the coordinate system to calculate the average brightness of the pixel points determined based on each width value point in the sub-region;

[0067] And the length data of each sub-region is used as the horizontal axis of the coordinate system to calculate the average brightness of the pixel points determined based on each length value point in the sub-region.

[0068] Specifically, the feature extraction module is used to draw brightness performance curves in different directional dimensions, where:

[0069] The feature extraction module is used to draw a first brightness performance curve in a first direction based on the average brightness of the pixel points determined by the width value points, and to draw a second brightness performance curve in a second direction based on the average brightness of the pixel points determined by the length value points;

[0070] The horizontal axis of the first brightness performance curve is the width value of the sub-region, and the vertical axis is the average brightness of the pixels. The horizontal axis of the second brightness performance curve is the length value of the sub-region, and the vertical axis is the average brightness of the pixels.

[0071] See also Figure 2As shown in FIG, it is a schematic diagram of the sub-region distribution of an embodiment of the present invention. For example, the reference image can be divided into 20 columns along the width direction and 20 rows along the length direction. For the sub-region of the 6th column along the width direction, the brightness average of the pixels in the 20 sub-regions of the 6th column is calculated, and the calculated brightness average is used as the brightness average corresponding to the width value points with a width value of the 6th column. Similarly, for the sub-region of the 13th row along the length direction, the brightness average of the pixels in the 20 sub-regions of the 13th row is calculated, and the calculated brightness average is used as the brightness average corresponding to the length value points with a length value of the 13th row.

[0072] As you can understand, the present invention divides the reference image into several rectangular sub-regions and performs brightness statistics on each sub-region along both the width and length directions. In the width direction, the average brightness of each column of pixels is calculated; in the length direction, the average brightness of each row of pixels is calculated. This two-dimensional analysis can capture the brightness variation patterns in different directions within the sub-region. A first brightness curve is plotted to reflect the brightness distribution characteristics in the width direction, while a second brightness curve is plotted to reflect the brightness distribution characteristics in the length direction.

[0073] Specifically, see Figure 3 As shown, it is a logic flow chart of determining the aggregation analysis area by the feature aggregation module according to an embodiment of the present invention. The feature aggregation module is used to determine the aggregation analysis area based on the location information matching condition, wherein:

[0074] The feature aggregation module is used to obtain position information of each sub-region where a contour boundary segment exists, including a row sequence determined along the length direction of the reference image and a column sequence determined along the width direction of the reference image according to the position information;

[0075] The feature aggregation module is used to determine an area consisting of several sub-areas that meet the location information matching condition as the aggregate analysis area, and not to perform analysis processing on sub-areas that do not meet the location information matching condition;

[0076] The aggregate analysis area includes at least two sub-areas, and the position information matching condition is that the row sequence or column sequence of each sub-area is consistent.

[0077] It can be understood that the contour boundary of the target object is usually composed of adjacent or continuously arranged sub-regions. The feature aggregation module defines the aggregation analysis area based on the matching conditions of row consistency or column consistency: if multiple sub-regions have the same row sequence or column sequence, they are divided into one aggregation analysis area. This strategy of aggregating sub-regions in the same row or column for analysis can effectively capture the linear extension characteristics of the contour boundary and improve the continuity of the contour boundary analysis.

[0078] Specifically, the feature aggregation module is used to determine the comparison of the brightness performance curves of each sub-region, wherein:

[0079] The feature aggregation module is used to determine a maximum brightness difference based on a brightness performance curve of each sub-region within the aggregate analysis region, so as to calculate a standard deviation of the maximum brightness difference of the sub-regions within the aggregate analysis region.

[0080] Specifically, the feature aggregation module is used to mark untrusted sub-regions, where:

[0081] The feature aggregation module is used to screen the aggregation analysis area corresponding to the maximum value of the standard deviation, and mark the sub-areas corresponding to the maximum value and the minimum value of the maximum brightness difference in the corresponding aggregation analysis area as untrusted sub-areas.

[0082] It can be understood that the present invention calculates the maximum brightness difference in the corresponding dimension for the brightness performance curve of each sub-region in the aggregation analysis area through the feature aggregation module. The difference reflects the severity of the brightness change in the sub-region. The standard deviation of the maximum brightness difference of all sub-regions in the aggregation area is calculated. The larger the standard deviation, the more significant the difference in brightness change of each sub-region in the aggregation analysis area, that is, the brightness characteristics of some sub-regions in the contour boundary of the same row or column are inconsistent with the overall trend. The aggregation analysis area with the largest standard deviation among all aggregation areas is screened out. The brightness feature consistency of the sub-regions in the aggregation analysis area is the worst, and may include contour blurring caused by complex lighting, resulting in abnormal identified contour boundary segments. The feature consistency quantified by standard deviation and the screening method of locating abnormal sub-regions with extreme values ​​can effectively distinguish reliable boundary segments from unreliable boundary segments in complex lighting scenes, thereby improving the robustness of overall contour extraction.

[0083] Specifically, the local processing module is used to obtain the brightness performance curves of the pixels in the non-confidence sub-region after adjusting the exposure in different directions, wherein:

[0084] The local processing module is configured to draw a third brightness performance curve in the first direction based on the average brightness of the pixels determined at each width value point in the non-confident sub-region after the exposure adjustment, and to draw a fourth brightness performance curve in the second direction based on the average brightness of the pixels determined at each length value point in the non-confident sub-region after the exposure adjustment;

[0085] The exposure of the non-confidence sub-region after adjustment is higher than the exposure before adjustment.

[0086] In the implementation of the present invention, those skilled in the art can adjust the exposure of the non-confidence sub-area of ​​the reference image according to processing requirements. Those skilled in the art can understand that increasing the exposure will magnify the difference between the boundary of the target object and the image background, but overexposure will also cause the loss of contour details of the target object. Based on this, in the implementation of the present invention, the exposure of the non-confidence sub-area can be adjusted to [103%, 110%] of the original exposure. Preferably, the exposure of the non-confidence sub-area can be adjusted to 105% of the original exposure.

[0087] It can be understood that by increasing the exposure of the marked non-confidence sub-region, the brightness difference of pixels in the region that was originally masked by underexposure can be amplified, especially the brightness gradient change between the target object and the background will be more obvious. After the exposure is enhanced, the module recalculates the average pixel brightness of each coordinate point in the width and length directions of the sub-region, and draws the third brightness performance curve and the fourth brightness performance curve respectively. Since the increase in exposure will cause the pixel value of the dark area to shift to the high brightness range, the brightness mutation feature at the originally blurred boundary will be more obvious, thereby realizing the highlighting of boundary features through local exposure enhancement.

[0088] Specifically, the local processing module is used to determine the contour boundary fitting points, wherein,

[0089] The local processing module is configured to compare the third brightness performance curve with the brightness average value of the first brightness performance curve on the vertical axis to determine a width value of a maximum difference in the brightness average value on the horizontal axis, and to compare the fourth brightness performance curve with the brightness average value of the second brightness performance curve on the vertical axis to determine a length value of the maximum difference in the brightness average value on the horizontal axis;

[0090] The local processing module is configured to determine the contour boundary fitting point in the non-confidence sub-region based on the width value and the length value.

[0091] It is understood that the present invention plots first and second luminance performance curves for non-confidence sub-regions before exposure adjustment, and then generates third and fourth luminance performance curves after exposure adjustment. The local processing module compares the curves before and after adjustment on the vertical axis, identifying the location with the largest difference between them. In the width direction, the horizontal axis width corresponding to the maximum difference between the average brightness of the third and first curves reflects the location with the most significant brightness change after exposure enhancement in that direction, typically corresponding to the boundary between the target and the background. Similarly, the length direction length corresponds to the maximum difference between the fourth and second curves. The point determined by the intersection of the determined width and length values ​​in the sub-region coordinate system is the contour boundary fitting point. Those skilled in the art will understand that exposure adjustment amplifies the brightness difference between the target and the background, and this difference is often more dramatic at the boundary than in non-boundary areas. By capturing the location of the extreme difference in the curve comparison, boundary points that were originally blurred due to lighting issues can be accurately located. This invention utilizes the complementary information before and after exposure adjustment, avoids the limitations of contour feature extraction, and achieves local optimization of local regions where local features are insufficient.

[0092] Specifically, the local processing module is used to fit the contour boundary segment of the non-confident sub-region, wherein:

[0093] The local processing module is used to obtain the initial contour boundary segment of the non-confidence sub-region, and fit the initial contour boundary segment based on the contour boundary fitting point, so that the fitted contour boundary segment passes through the contour boundary fitting point.

[0094] In the implementation of the present invention, the initial contour boundary segments of the non-confidence sub-region obtained by the preprocessing module can be uniformly sampled in advance to generate a discrete point set, the determined contour boundary fitting points are added to the discrete point set, and then Gaussian filtering smoothing fitting is performed based on the points in the discrete point set to obtain the completed fitted contour boundary segments. Fitting curves based on discrete points is an existing technology and will not be repeated here.

[0095] Specifically, the output end is used to update the first contour boundary using the contour boundary segment that has completed fitting, wherein,

[0096] The output end is used to replace the initial contour boundary segment with the contour boundary segment that has completed fitting, so as to update the first contour boundary to the second contour boundary.

[0097] It can be understood that, compared with the first contour boundary, the contour boundary segments in the non-confidence sub-region of the second contour boundary are re-fitted and replaced, and the second contour boundary is the final contour boundary of the target object determined for the reference image.

[0098] Thus far, the technical solutions of the present invention have been described in conjunction with the preferred embodiments shown in the accompanying drawings. However, it will be readily understood by those skilled in the art that the scope of protection of the present invention is obviously not limited to these specific embodiments. Without departing from the principles of the present invention, those skilled in the art may make equivalent changes or substitutions to the relevant technical features, and the technical solutions after such changes or substitutions will fall within the scope of protection of the present invention.

[0099] The foregoing description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. Those skilled in the art will readily appreciate that the present invention is susceptible to various modifications and variations. Any modifications, equivalent substitutions, or improvements made within the spirit and principles of the present invention shall be included within the scope of protection of the present invention.

Claims

1. A computer image processing system based on deep learning, characterized in that: include: a preprocessing module configured to divide the reference image into a plurality of rectangular sub-regions, determine a first contour boundary of the target object based on a deep learning model, and mark position information of the sub-regions where contour boundary segments exist; A feature extraction module, which is connected to the preprocessing module and is used to obtain the brightness of the pixels in each sub-region to draw brightness performance curves of the pixels in each sub-region in different directions and dimensions; a feature aggregation module, connected to the preprocessing module and the feature extraction module, respectively, for determining an aggregate analysis area based on position information matching conditions, and marking untrusted sub-areas according to the comparison of brightness performance curves of each sub-area within the aggregate analysis area; a local processing module connected to the feature aggregation module, configured to obtain brightness performance curves of the pixels of the non-confident sub-region in different directions after the exposure is adjusted, and determine contour boundary fitting points based on comparison of the brightness performance curves after the exposure is adjusted, so as to fit the contour boundary segments of the non-confident sub-region; An output end is connected to the local processing module and the pre-processing module respectively, and is used to update the first contour boundary using the contour boundary segment of the non-confidence sub-region to output a second contour boundary of the target object in the reference image.

2. The computer image processing system based on deep learning according to claim 1, characterized in that The feature extraction module is used to calculate the average brightness of the pixels in each sub-region in different directions, where: The feature extraction module is used to use the width data of each sub-region as the horizontal axis of the coordinate system to calculate the average brightness of the pixel points determined based on each width value point in the sub-region; And the length data of each sub-region is used as the horizontal axis of the coordinate system to calculate the average brightness of the pixel points determined based on each length value point in the sub-region.

3. The computer image processing system based on deep learning according to claim 2, characterized in that The feature extraction module is used to draw brightness performance curves in different directional dimensions, wherein: The feature extraction module is used to draw a first brightness performance curve in a first direction based on the average brightness of the pixel points determined by the width value points, and to draw a second brightness performance curve in a second direction based on the average brightness of the pixel points determined by the length value points; The horizontal axis of the first brightness performance curve is the width value of the sub-region, and the vertical axis is the average brightness of the pixels. The horizontal axis of the second brightness performance curve is the length value of the sub-region, and the vertical axis is the average brightness of the pixels.

4. The computer image processing system based on deep learning according to claim 3, characterized in that The feature aggregation module is used to determine the aggregation analysis area based on the location information matching condition, wherein: The feature aggregation module is used to obtain position information of each sub-region where a contour boundary segment exists, including a row sequence determined along the length direction of the reference image and a column sequence determined along the width direction of the reference image according to the position information; The feature aggregation module is used to determine an area composed of several sub-areas that meet the location information matching condition as the aggregation analysis area; The aggregate analysis area includes at least two sub-areas, and the position information matching condition is that the row sequence or column sequence of each sub-area is consistent.

5. The computer image processing system based on deep learning according to claim 4, characterized in that The feature aggregation module is used to determine the comparison of the brightness performance curves of each sub-region, wherein: The feature aggregation module is used to determine a maximum brightness difference based on a brightness performance curve of each sub-region within the aggregate analysis region, so as to calculate a standard deviation of the maximum brightness difference of the sub-regions within the aggregate analysis region.

6. The computer image processing system based on deep learning according to claim 5, characterized in that The feature aggregation module is used to mark non-confident sub-regions, wherein: The feature aggregation module is used to screen the aggregation analysis area corresponding to the maximum value of the standard deviation, and mark the sub-areas corresponding to the maximum value and the minimum value of the maximum brightness difference in the corresponding aggregation analysis area as untrusted sub-areas.

7. The computer image processing system based on deep learning according to claim 6, characterized in that The local processing module is used to obtain the brightness performance curves of the pixels in the non-confidence sub-region after the exposure is adjusted in different directions, wherein: The local processing module is configured to draw a third brightness performance curve in the first direction based on the average brightness of the pixels determined at each width value point in the non-confident sub-region after the exposure adjustment, and to draw a fourth brightness performance curve in the second direction based on the average brightness of the pixels determined at each length value point in the non-confident sub-region after the exposure adjustment; The exposure of the non-confidence sub-region after adjustment is higher than the exposure before adjustment.

8. The computer image processing system based on deep learning according to claim 7, characterized in that The local processing module is used to determine the contour boundary fitting points, wherein, The local processing module is configured to compare the third brightness performance curve with the brightness average value of the first brightness performance curve on the vertical axis to determine a width value of a maximum difference in the brightness average value on the horizontal axis, and to compare the fourth brightness performance curve with the brightness average value of the second brightness performance curve on the vertical axis to determine a length value of the maximum difference in the brightness average value on the horizontal axis; The local processing module is configured to determine the contour boundary fitting point in the non-confidence sub-region based on the width value and the length value.

9. The computer image processing system based on deep learning according to claim 8, characterized in that The local processing module is used to fit the contour boundary segment of the non-confident sub-region, wherein: The local processing module is used to obtain the initial contour boundary segment of the non-confidence sub-region, and fit the initial contour boundary segment based on the contour boundary fitting point, so that the fitted contour boundary segment passes through the contour boundary fitting point.

10. The computer image processing system based on deep learning according to claim 9, characterized in that The output end is used to update the first contour boundary using the contour boundary segment that has completed fitting, wherein, The output end is used to replace the initial contour boundary segment with the contour boundary segment that has completed fitting, so as to update the first contour boundary to the second contour boundary.

Images