A visual image processing method based on brain-inspired processing

Abstract

A visual image processing method based on brain-like processing, belonging to the technical field of image recognition, includes the following steps: Step 1. Divide the image into multiple rows and columns in units of pixels, scan row by row, take the pixel value of each point as the ordinate and the column number where it is located as the abscissa to obtain the digital visual wave of this row. Step 2. Screen out the bands to be screened in each row; Step 3. Take the center points, as well as the front and back endpoints of each band to be screened for screening. Step 32. Perform graphic fitting recognition on the center points and endpoints of the bands to be screened retained in each row and column to obtain the image. The present invention directly extracts, infers, fits and clusters the image pixel value change points through row-by-row scanning and calculation to obtain and recognize the image, improves the image recognition speed and accuracy, reduces the hardware requirements, and is suitable for real-time automatic search, recognition and tracking applications of moving and weak targets in various complex environments.

Description

Technical Field

[0001] The present invention belongs to the field of image recognition, and particularly relates to a visual image processing method based on brain-like processing. Background Art

[0002] How vision captures the things and objects of concern from the complex, ever-changing, intertwined, and mutually restrictive all things is a major problem in visual artificial intelligence. Especially for the motion scenes of objects in various complex environments, it is even more difficult to capture the objects of concern well, seriously hindering the development of the urgently needed visual artificial intelligence industry.

[0003] At present, people adopt the method of obtaining a large number of visual scene samples and conducting in-depth training and learning to solve this problem. However, it is very difficult to obtain relatively complete training and learning samples in the rapidly changing motion scenes. Therefore, it can only be applied to fixed or slightly changing scenes and still has certain limitations.

[0004] For the recognition of low-altitude target aircraft against the sky background, due to being at low altitude, conventional radars are prone to have dead angles. At this time, image recognition technology can be used for low-altitude recognition. Due to the high speed in the flight area, whether tracking the target or capturing the target, a very fast reaction speed is required. Using the deep learning sampling method for image recognition has a huge amount of calculation, and the loop is complex and prone to cause program deadlocks, with a slow reaction speed, and the actual application effect in low-altitude target tracking is not good. Summary of the Invention

[0005] To overcome the technical defects existing in the prior art, the present invention discloses a visual image processing method based on brain-like processing.

[0006] The visual image processing method based on brain-like processing according to the present invention includes the following steps:

[0007] Step 1. Divide the image into multiple rows and columns by pixels, and scan each row row by row. Using the pixel value of each point in the row as the ordinate and the column number where the point is located as the abscissa, obtain the digital visual wave of the row;

[0008] Step 2. Set the first merging threshold N1 and the first adjacent difference threshold C1; screen out the bands to be screened in each row;

[0009] The judgment method for marking each band in the digital visual wave as a non-screening band is: the difference between the abscissas of all points in the band and the starting point of the band does not reach the first merging threshold N1, and the difference between the maximum and minimum values of the ordinates of all points in the band does not reach the first adjacent difference threshold C1;

[0010] The bands composed of the remaining points are marked as bands to be screened;

[0011] Scan the picture line by line to obtain the bands to be screened for each line, and scan column by column to obtain the bands to be screened for each column;

[0012] Step 3. Perform regional fitting on the bands to be screened and output the recognition image. Specifically:

[0013] Step 31. Take the center point, as well as the front and rear end points, of each band to be screened, and set the second merging threshold N2. If the difference between the front and rear end points is less than the second merging threshold N2, represent the band to be screened directly by the center point;

[0014] If the difference between the front and rear end points is not less than the second merging threshold N2, represent the band to be screened by the front and rear end points;

[0015] Step 32. Perform graphic fitting recognition on the center points and end points of the bands to be screened retained for each row and each column to obtain an image.

[0016] Preferably, the specific method for obtaining the bands to be screened in step 2 is:

[0017] Step 21. Read the pixel values of the first point and the second point in the digital visual wave, and set the smaller of the pixel values of the two points as the minimum value and the other as the maximum value;

[0018] Step 22. Compare each point from the third point to the N1th point with the minimum value and the maximum value respectively; if it is less than the minimum value, update the minimum value with the pixel value of this point; if it is greater than the maximum value, update the maximum value with the pixel value of this point; in other cases, do not update;

[0019] When step 22 reaches the N1 + 1th point, enter step 23;

[0020] Step 23. Start from the jth point, where j = N1 + 1, N1 + 2…;

[0021] Define the maximum value and the minimum value of the jth point as the maximum value and the minimum value respectively among the N1 vertical coordinates of all points from the j - N1th point to the jth point;

[0022] Calculate the difference between the maximum value and the minimum value of the jth point. If it is greater than the first adjacent difference threshold C1, start marking from the jth point as a band to be screened and continue to the next j + 1th point; until the absolute value of the difference between the maximum value and the minimum value of a certain point is not greater than the first adjacent difference threshold C1, the band to be screened ends;

[0023] Step 24. Mark the remaining points that are not bands to be screened as non - screening bands.

[0024] Preferably, in step 2, after step 24, it further includes backtracking screening. Specifically:

[0025] Starting from the second line, compare the number of columns of the bands to be screened, i.e., the ordinate, in the Kth line with that in the previous line, i.e., the (K - 1)th line, where K is greater than or equal to 2.

[0026] Step 251. First, determine whether there is partial overlap between a certain band to be screened SCi in the (K - 1)th line and a certain band to be screened in the Kth line. If so, retain the band to be screened SCi in the (K - 1)th line; at the same time, mark the band to be screened SCj1 that overlaps with the band to be screened SCi in the Kth line and the (K - 1)th line as a band to be screened that has been inspected.

[0027] Otherwise, proceed to Step 252.

[0028] Step 252. Continue to check the difference in the ordinate of the endpoints of the band to be screened SCi at both ends of the (K - 1)th line and the endpoints of the nearest band to be screened in the Kth line. When the difference in the ordinate does not exceed the set column threshold LT, retain the band to be screened SCi in the (K - 1)th line, and mark the nearest band to be screened SCj2 in the Kth line to the band to be screened SCi as a band to be screened that has been inspected; when the difference in the ordinate exceeds the set column threshold LT, re-mark the band to be screened SCi in the (K - 1)th line as a band not to be screened, and do not allow it to enter the subsequent steps.

[0029] Step 253. Traverse all the bands to be screened in the (K - 1)th line.

[0030] Step 254. Starting from the third line, repeat the above Steps 251 to 253 for the remaining bands to be screened in the previous line except for the bands to be screened that have been inspected.

[0031] Preferably, in Step 2, after Step 24, it further includes backtracking screening, specifically:

[0032] Starting from the Kth line, where K is greater than or equal to M, and M is the defined minimum row span.

[0033] Define the adjacent row detection process, including the following steps:

[0034] Step 251. First, determine whether there is partial overlap between a certain band to be screened SCi in the (K - 1)th line and a certain band to be screened in the Kth line. If so, retain the band to be screened SCi in the (K - 1)th line; at the same time, mark the band to be screened SCj1 that overlaps with the band to be screened SCi in the Kth line and the (K - 1)th line as a band to be screened that has been inspected.

[0035] Otherwise, proceed to Step 252.

[0036] Step 252. Continue to check the difference in the vertical coordinates of the two endpoints of the to-be-screened band SCi in the (K - 1)-th row and the endpoints of the to-be-screened band closest to it in the K-th row. When the difference in the vertical coordinates does not exceed the set column threshold LT, retain the to-be-screened band SCi in the (K - 1)-th row, and mark the to-be-screened band SCj2 closest to the to-be-screened band SCi in the K-th row as the screened to-be-screened band; when the difference in the vertical coordinates exceeds the set column threshold LT, re-mark the to-be-screened band in the (K - 1)-th row as a non-screening band and do not let it enter the subsequent steps;

[0037] Step 253. Traverse all the to-be-screened bands in the (K - 1)-th row;

[0038] The backtracking screening for a to-be-screened band SCi1 in the K-th row specifically includes the following steps:

[0039] Start with a to-be-screened band SCi1 in the (K - 1)-th row; perform the adjacent row detection process described in Steps 251 to 253;

[0040] If no screened to-be-screened band can be obtained in the K-th row, re-mark the to-be-screened band SCi1 in the (K - 1)-th row as a non-screening band and terminate the backtracking screening for the to-be-screened band SCi1;

[0041] If a screened to-be-screened band is obtained in the K-th row, mark the to-be-screened band SCi1 in the (K - 1)-th row as the backtracking target;

[0042] For all the to-be-screened bands in the (K - 2)-th row, continue with the following Step 255;

[0043] Step 255 is specifically as follows: Perform the adjacent row detection process described in Steps 251 to 253, and determine whether there is a to-be-screened band in this row, i.e., the (K - 2)-th row, whose corresponding screened to-be-screened band is the backtracking target;

[0044] If not, re-mark the to-be-screened band SCi1 in the K-th row as a non-screening band;

[0045] If there is such a to-be-screened band, mark this to-be-screened band SCi2 in this row, i.e., the (K - 2)-th row, as the new backtracking target;

[0046] For the (K - 3)-th row to the (K - M + 1)-th row, repeat Step 255 row by row; determine whether there is a to-be-screened band in this row whose corresponding screened to-be-screened band is the backtracking target;

[0047] If there is none in any row, re-mark the to-be-screened band SCi1 in the K-th row as a non-screening band and terminate the backtracking screening;

[0048] If there is such a band to be screened, mark this band to be screened in this line as the new backtracking target, and go back to the previous line to perform step 255 again until the (K - M + 1)-th line.

[0049] Preferably, step 32 is specifically as follows:

[0050] Step 321. Take the center points, as well as the front and back end points of each band to be screened as the point library of points to be clustered.

[0051] Step 322. Arbitrarily select a point to be clustered in the point library of points to be clustered as the target point, and find all adjacent points to be clustered of this clustered point in the point library of points to be clustered. The point library of points to be clustered is the point library composed of the reserved center points and end points of the bands to be screened.

[0052] The adjacent points to be clustered are the points whose spatial distance from the target point is less than the spatial threshold.

[0053] Step 323. Remove the selected adjacent points to be clustered from the point library of points to be clustered; for each selected adjacent point to be clustered, take it as the target point, and continue to find all adjacent points to be clustered of this target point in the point library of points to be clustered.

[0054] Step 324. Repeat step 323 until no adjacent points to be clustered can be found in the point library of points to be clustered, and one clustering is completed.

[0055] Step 325. Repeat steps 322 to 324 until all points in the point library of points to be clustered are clustered.

[0056] Each clustering results in a cluster composed of multiple points to be clustered.

[0057] Step 326. Perform the above-mentioned graphic fitting recognition on the points to be clustered within each cluster.

[0058] Preferably, step 326 is specifically as follows:

[0059] Step 3261. Classify directly according to the number of adjacent points to be clustered owned. Mark the points to be clustered with more than 3 adjacent points to be clustered as central points, the points to be clustered with 2 adjacent points to be clustered as connection points, and the points to be clustered with 1 adjacent point to be clustered as end points.

[0060] Step 3262. According to the connection relationship, for all central points, mark all the connection points connecting this central point to other central points or end points as one group, and perform fitting on all the points to be clustered within one group; if among the three groups of points to be clustered connected by a certain central point, there are two groups of non-coincident straight lines fitted, it indicates that there is a broken line.

[0061] Step 3263. Prioritize the recognition of clusters with broken lines, that is, recognize the cluster where the central point is located; if there are no broken lines in each cluster, directly recognize each cluster in sequence.

[0062] Preferably, after obtaining all the clusters in step 325, it further includes gray-scale joint correction for the clusters, specifically:

[0063] For each row, mark each segment separated by the to-be-screened difference band as a merged segment, and calculate the pixel average value within each merged segment as the pixel mean value of the merged segment;

[0064] Set the second adjacent difference threshold C2. Between adjacent rows, the merged segments with overlapping column coordinates are used as adjacent merged segments between rows. If the difference in pixel mean values between two adjacent merged segments between different rows is less than the second adjacent difference threshold C2, then the to-be-clustered points corresponding to the to-be-screened difference bands on both sides of these two adjacent merged segments between rows are regarded as the same cluster, and those greater than the second adjacent difference threshold are not regarded as the same cluster.

[0065] Preferably, in step 326 for all the clusters after clustering, traverse each cluster, judge the positions of all the clustered points within each cluster, and judge whether there are clustered points located at the image boundary. If not, prioritize the in-depth recognition of this cluster.

[0066] Preferably, in step 2, the non-screening bands are merged, and the ordinate of all points within the merged band is taken as the average value of the ordinates of all points within the band.

[0067] Adopting the visual image processing method based on brain-like processing of the present invention, by simulating the rapid focusing recognition of pixel mutation points in the image by the human brain, directly identifying the boundary of pixel point difference change to extract endpoints and cluster, and recognizing the clusters after clustering, converting the overall recognition of a large image into the recognition of small contour lines, improving the image recognition speed and reducing the hardware requirements for image recognition, which is suitable for the application of aircraft recognition in the sky background with relatively pure background and high requirements for recognition speed. Brief Description of the Drawings

[0068] Figure 1 It is a schematic flow chart of a specific implementation manner of the image recognition method of the present invention;

[0069] Figure 2 It is a schematic flow chart of a specific implementation manner of step 2 of the present invention;

[0070] Figure 3 It is a schematic flow chart of a specific implementation manner of step 3 of the present invention;

[0071] Figure 4Schematic diagram of a specific implementation manner of the backtracking screening described in the present invention;

[0072] Figure 5 Schematic diagram of another specific implementation manner of the backtracking screening described in the present invention;

[0073] Figure 6 Schematic diagram of a specific implementation manner of a cluster obtained after clustering once in the present invention;

[0074] Figure 7 Schematic diagram of a specific implementation manner of the grayscale region combination described in the present invention;

[0075] Figure 8 Schematic diagram of a specific implementation manner of the digital vision wave described in the present invention;

[0076] Figure 9 Schematic diagram of a specific implementation manner of the recognition method described in the present invention for recognition;

[0077] In the figure, the names of the reference numerals are: 1 - endpoint, 2 - connection point, SC - band to be screened. Specific implementation manner

[0078] The following further elaborates on the specific implementation manner of the present invention in conjunction with the accompanying drawings.

[0079] The visual image processing method based on brain-like processing described in the present invention includes the following steps:

[0080] Step 1. Divide the image into multiple rows and columns, and scan each row row by row. Using the pixel value of each point in the row as the ordinate and the column number where the point is located as the abscissa, obtain the digital vision wave of the row; as Figure 8 shown, Figure 8 the upper part in the figure is an image based on a sky background, and the lower part is Figure 8 the digital vision wave of a certain row in the figure, with the abscissa being the column number and the ordinate being the pixel value.

[0081] Step 2. Set the first merging threshold N1 and the first adjacent difference threshold C1; screen out the bands to be screened in each row;

[0082] The method for determining that each band in the digital vision wave is a non-screening band is: the difference between the abscissas of all points in the band and the starting point of the band does not reach the first merging threshold N1, and the difference between the maximum and minimum values of the ordinates of all points in the band does not reach the first adjacent difference threshold C1;

[0083] After typical digital vision waves are marked, a combination of several non-screening bands and bands to be screened with different ordinates is obtained;

[0084] For example, set the first merging threshold N1 = 5 and the first adjacent difference threshold C1 = 10. For 20 pixel points arranged from left to right in a certain row, the abscissa and ordinate are respectively:

[0085] (1, 32), (2, 34), (3, 35), (4, 36), (5, 38), (6, 41), (7, 41), (8, 43), (9, 44), (10, 47), (11, 50), (12, 54), (13, 57), (14, 58), (15, 60), (16, 61), (17, 62), (18, 63), (19, 64), (20, 66);

[0086] For the absolute value of the difference in the ordinate of each point from the 1st to the 7th point, there is no case reaching the first adjacent difference threshold C1 = 10. Although there is a case where the value is greater than the first adjacent difference threshold C1 = 10 for the 8th point when looking forward, the minimum ordinate value of 32 appears at the 1st point, and the difference in the abscissa from the 8th point is greater than the first merging threshold of 5;

[0087] Similar conclusions can be drawn for the 9th, 10th, and 11th points as for the 8th point;

[0088] For the 12th point, the difference in the ordinate from the 8th point forward is 54 - 43 = 11, which exceeds the first adjacent difference threshold C1 = 10, and the difference in their abscissas is 4, which does not reach the first merging threshold N1 = 5. That is, from the analysis of the 12th point, it can be obtained that the 8th to 12th points meet the marking conditions for the band to be screened;

[0089] Similar conclusions can be drawn for the 13th, 14th, and 15th points as for the 12th point. The minimum ordinate points corresponding are the 9th, 10th, and 11th points respectively;

[0090] That is, the 8th to 15th points meet the marking conditions for the band to be screened;

[0091] For the 16th point, among the points with an abscissa difference less than 5 when looking forward, the ordinate of the 12th point is 54, and the difference from the ordinate of the 16th point of 61 does not reach the first adjacent difference threshold C1 = 10, meeting the marking conditions for the non-screening band;

[0092] Similar conclusions can also be drawn for the 17th, 18th, 19th, and 20th points.

[0093] By comparing each point forward, it can be obtained that the above 20 points can be divided into 3 segments. The 1st to 7th points are the non-screening band, the 8th to 15th points are the band to be screened, and the 16th to 20th points are the non-screening band.

[0094] Through marking, it is possible to obtain that the non-screening band is the band where there is no drastic pixel change, and the band to be screened is the band where there is a drastic pixel change. By setting the first merging threshold N1, it is possible to effectively avoid noise signals or sudden changes in pixel values at individual points due to imaging reasons, and screen out the gradually changing but significantly changing pixel change regions.

[0095] A subsequent typical processing method for multiple points in the same non-screening band with continuous abscissas is to perform merging, where the band with the largest difference in the starting abscissa points is selected for merging; for all points within the merged band, the ordinate of each point is taken as the average of the ordinates of all points within the band.

[0096] After merging, during the subsequent image recognition process, each non-screening band can be ignored during image recognition. Since the non-screening band is the area where there is no drastic pixel change and is not the boundary of the pattern in the image, it has no impact on the image recognition process. Ignoring it can significantly reduce the computational complexity during the image recognition process.

[0097] An operation method that is conducive to software programming implementation for realizing the above marking and judgment process is as follows:

[0098] Step 21. Read the pixel values of the first point and the second point in the digital visual wave, set the smaller one as the minimum value and the larger one as the maximum value.

[0099] Step 22. Compare each point from the third point to the N1th point with the minimum value and the maximum value respectively; if it is less than the minimum value, update the minimum value with the pixel value of this point; if it is greater than the maximum value, update the maximum value with the pixel value of this point; in other cases, do not update.

[0100] When step 22 reaches the (N1 + 1)th point, enter step 23.

[0101] Step 23. Start from the jth point, where j = N1 + 1, N1 + 2...

[0102] Define the maximum value and the minimum value of the jth point as the maximum value and the minimum value respectively among the ordinates of all N1 points from the (j - N1)th point to the jth point.

[0103] Calculate the difference between the maximum value and the minimum value of the jth point. If it is greater than the first adjacent difference threshold C1, start marking as the band to be screened from the jth point; until the absolute value of the difference between the maximum value and the minimum value of this point is not greater than the first adjacent difference threshold C1, the band to be screened ends.

[0104] For example, set the first merging threshold N1 = 5 and the first adjacent difference threshold C1 = 10. For 20 pixel points arranged from left to right in a certain row, the abscissa and ordinate are respectively:

[0105] (1, 32), (2, 34), (3, 35), (4, 36), (5, 38), (6, 41), (7, 41), (8, 43), (9, 44), (10, 47), (11, 50), (12, 54), (13, 57), (14, 58), (15, 60), (16, 61), (17, 62), (18, 63), (19, 64), (20, 66);

[0106] If the pixel values of the first point and the second point are 32 and 34 respectively, then set 32 as the minimum value and 34 as the maximum value;

[0107] The pixel value of the third point is 35, which is greater than the maximum value. Then update the maximum value to 35. And so on, when updating to the N1-th point, that is, the fifth point, the updated maximum value is 38 and the minimum value is 32;

[0108] For the first N1 points, since there cannot be a situation where the difference is greater than the merging threshold, and they appear at the image boundary. Pixel mutations mostly occur due to shooting or imaging reasons. Even if the reason for the pixel mutation is the existence of the target to be recognized, it can be discovered through the recognition of subsequent points. The recognition significance of pixel mutations only occurring in the first N1 points is extremely limited and is generally ignored.

[0109] Starting from the (N1 + 1)-th point, calculate the difference between the maximum value and the minimum value. If it is greater than the first adjacent difference threshold C1 = 10, it indicates the occurrence of a band to be screened;

[0110] For example, the pixel value of the sixth point is 41, the minimum value is 32, and the difference between them is 8, which is less than 10, not meeting the requirements for marking the band to be screened;

[0111] For the eighth point (8, 43), the pixel value is 43. Among all the points from the eighth point to the j - N1-th point, that is, the third point, the maximum value is 43, the minimum value is 35, and the difference is 8, which is less than 10, not meeting the requirements for marking the band to be screened;

[0112] For the eleventh point (11, 50), the pixel value is 50. Among all the points from the eleventh point to the j - N1-th point, that is, the sixth point, the maximum value is 50, the minimum value is 41, and the difference is 9, which is less than 10, not meeting the requirements for marking the band to be screened;

[0113] For the twelfth point (12, 54), the pixel value is 54. Among all the points from the twelfth point to the j - N1-th point, that is, the seventh point, the maximum value is 54, the minimum value is 41, and the difference is 13, which is greater than 10, meeting the requirements for marking the band to be screened.

[0114] In this step, the band to be screened obtained is the area where pixel values change significantly in each row; it represents the contour boundary in the picture that has a large difference from the surrounding pixel values. For example, it may be clouds, birds, aircraft, etc. in the sky.

[0115] Scan the image line by line to obtain the bands to be screened for each line, and then scan column by column to obtain the bands to be screened for each column;

[0116] To further reduce the workload in subsequent graphic fitting, backtracking screening can be performed during the generation of the bands to be screened. Specifically:

[0117] Starting from the second row, compare the number of columns (i.e., the ordinate) of the bands to be screened in the Kth row with that in the row above it, i.e., the (K - 1)th row, where K is greater than or equal to 2;

[0118] Step 251. First, determine whether there is partial overlap between a certain band to be screened SCi in the (K - 1)th row and a certain band to be screened in the Kth row. If so, retain the band to be screened in the (K - 1)th row; at the same time, mark the band to be screened SCj1 in the Kth row that overlaps with the band to be screened SCi in the (K - 1)th row as the band to be screened that has been inspected;

[0119] Otherwise, go to Step 252.

[0120] Step 252. Continue to check the difference in the ordinate of the two endpoints of the band to be screened SCi in the (K - 1)th row and the endpoints of the nearest band to be screened in the Kth row. When the difference in the ordinate does not exceed the set column threshold LT, retain the band to be screened SCi in the (K - 1)th row, and mark the nearest band to be screened SCj2 in the Kth row to the band to be screened SCi as the band to be screened that has been inspected; when the difference in the ordinate exceeds the set column threshold LT, re - mark the band to be screened SCi in the (K - 1)th row as the band not to be screened, and do not let it enter the subsequent steps.

[0121] As Figure 4 shown, if there is no band to be screened SCj1 in the (K - 1)th row that overlaps with the band to be screened SCi in the Kth row, then detect the difference in the ordinate of the nearest band to be screened SCj2 to the band to be screened SCi. When the difference in the ordinate does not exceed the set column threshold LT, retain the band to be screened SCi in the (K - 1)th row, and mark the nearest band to be screened SCj2 in the Kth row to the band to be screened SCi as the band to be screened that has been inspected; when the difference in the ordinate exceeds the set column threshold LT, re - mark the band to be screened in the (K - 1)th row as the band not to be screened, and do not let it enter the subsequent steps.

[0122] Step 253. Traverse all the bands to be screened in the (K - 1)th row.

[0123] Step 254. Starting from the third row, repeat the above Steps 251 to 253 for the remaining bands to be screened in the previous row except for the bands to be screened that have been inspected.

[0124] The above backtracking screening process can be carried out synchronously when screening the bands to be screened for each row from top to bottom in Step 2.

[0125] The complete contours in the image, especially the contours of the aircraft, are necessarily distributed continuously across multiple rows. If there is no other screening band associated with the screening band to be screened in adjacent rows for a certain screening band to be screened, it indicates that the screening band to be screened is not part of a complete contour, usually generated by noise or natural light, and no subsequent graphic fitting is required.

[0126] To determine whether there is an association, it is judged according to the set inter-column threshold LT. For example, if LT1 = 4 is set, it is considered that the screening bands to be screened with an interval of no more than 4 between adjacent rows are all associated with the screening band S1.

[0127] By using a one-way comparison from top to bottom row by row upwards, the screening bands to be screened in the upper row that have no association with the lower row can be excluded. Each screening band to be screened may be associated with multiple screening bands in the upper row, that is, each screening band to be screened may generate multiple screened bands to be screened; by retaining the screened bands to be screened, it is possible to avoid misdeleting the screening bands in the upper row that are associated with the row above the upper row, and it can be carried out synchronously when screening the screening bands row by row, improving the running speed.

[0128] According to the above idea, assuming that at least M rows need to be spanned for the contour that can be recognized as an aircraft during the subsequent recognition process, after the screening bands to be screened from the first to the (M - 1)th row are generated, the backtracking screening starts from the Mth row. The purpose of the backtracking screening is to exclude those screening bands to be screened that continuously do not reach M rows, thereby reducing the workload of subsequent steps.

[0129] Starting from the Kth row, compare the ordinate of the screening bands to be screened in this Kth row with those in the M - 1 rows above this row; K is greater than or equal to M;

[0130] Define the steps 251 - 253 described above as an adjacent row detection process.

[0131] As Figure 5 shown, the backtracking screening of a certain screening band SCi1 in the Kth row specifically includes the following steps:

[0132] Start from a certain screening band SCi1 in the (K - 1)th row; perform the adjacent row detection process described in steps 251 to 253;

[0133] If no screened band to be screened can be obtained in the Kth row, for example, no screened bands such as SCj1 or SCj2 can be obtained, then the screening band SCi1 in the (K - 1)th row is re-labeled as a non-screening band, and the backtracking screening for the screening band SCi1 is terminated;

[0134] If a screened band to be screened is obtained in the Kth row, then mark the screening band SCi1 in the (K - 1)th row as the backtracking target;

[0135] For all the bands to be screened in the (K - 2)-th row, continue with the following step 255;

[0136] Specifically, step 255 is as follows: Perform the adjacent row detection process described in steps 251 to 253, and determine whether there is a band to be screened in this row, i.e., the (K - 2)-th row, whose corresponding screened band to be screened is the backtracking target;

[0137] If not, re - mark the band to be screened SCi1 in the K - th row as a non - screening band;

[0138] If there is such a band to be screened, then mark this band to be screened SCi2 in this row, i.e., the (K - 2)-th row, as the new backtracking target;

[0139] For the (K - 3)-th row to the (K - M + 1)-th row, repeat step 255 row by row;

[0140] That is, for the (K - 2)-th row to the (K - M + 1)-th row, perform the adjacent row detection process described in steps 251 to 253 row by row, and determine whether there is a band to be screened in this row whose corresponding screened band to be screened is the backtracking target;

[0141] If there is none in any row, re - mark the band to be screened SCi1 in the K - th row as a non - screening band and terminate the backtracking screening;

[0142] If there is such a band to be screened, then mark this band to be screened in this row as the new backtracking target and enter the previous row to perform step 255 again.

[0143] Through step 255, those bands to be screened that continuously exist for less than M rows can be excluded, marked as non - screening bands, and not enter the subsequent steps.

[0144] Step 3. Extract the endpoints and center - point region fitting for the bands to be screened, and the recognized image can be directly output;

[0145] The bands to be screened actually represent the boundary contours of pixel blocks with large pixel differences in the image. Through row - column extraction, multiple boundary contours are obtained, and other pixel points are directly merged and ignored, thereby improving the recognition speed.

[0146] A specific implementation manner of step 3 is as follows:

[0147] Step 31. Take the center points, as well as the front and back endpoints, of each band to be screened, and set the second merging threshold N2. If the difference in the abscissas of the front and back endpoints is less than the second merging threshold N2, it indicates that the band to be screened is relatively narrow, approaching a point visible to the naked eye in the picture rather than a relatively wide area with continuous gray - scale changes to the naked eye. At this time, represent this band to be screened directly with the center point;

[0148] The band to be screened is relatively wide, indicating that there are large regions of continuous pixel variation in the image, such as clouds and areas where the aircraft faces the sun. Since the surfaces are approximately spherical, the reflection angles of sunlight are different, resulting in continuous pixel variation after imaging. The endpoints of the band to be screened are taken for fitting.

[0149] Step 32. Perform regional fitting on the center points and endpoints of the bands to be screened retained in each row and column, and the recognized image can be directly output.

[0150] To further shorten the recognition time, in Step 32, clustering can be preferentially performed on the center points and endpoints of the bands to be screened retained in each row and column. The clustering method is as follows:

[0151] Step 321. Take the center points, as well as the front and back endpoints of each band to be screened as the point library to be clustered.

[0152] Step 322. Arbitrarily select a point to be clustered in the point library to be clustered as the target point, and find all adjacent points to be clustered of this clustering point in the point library to be clustered. The point library to be clustered is the point library composed of the retained center points and endpoints of the bands to be screened;

[0153] The adjacent points to be clustered are points with a spatial distance less than the spatial threshold from the target point;

[0154] For example, the spatial coordinates of the target point are (h0, l0), where h0 and l0 are the row and column numbers of the target point respectively;

[0155] The spatial coordinates of a certain point to be clustered are (h1, l1), where h1 and l1 are the row and column numbers of this point to be clustered respectively;

[0156] The spatial distance L between the two is L = [(h1 - h0) 2 +(l1 - l0) 2 1 / 2 ; if L ≤ LM, then determine that this point to be clustered is an adjacent point to be clustered of the target point, where LM is the set spatial threshold;

[0157] Step 323. Remove the selected adjacent points to be clustered from the point library to be clustered; for each selected adjacent point to be clustered, use it as the target point and continue to find all adjacent points to be clustered of this target point in the point library to be clustered;

[0158] Step 324. Repeat Step 323 until no adjacent points to be clustered can be found in the point library to be clustered, and one clustering is completed.

[0159] Step 325. Repeat Steps 322 to 324 until all points in the point library to be clustered are clustered.

[0160] ​Through the above method, each clustering results in a cluster composed of multiple points to be clustered, and the points to be clustered within the cluster are continuously distributed in space;

[0161] Step 326. Perform graphic fitting on the points to be clustered within each cluster.

[0162] To accelerate the recognition speed, it is necessary to classify and recognize the points to be clustered within each cluster. As Figure 6 shown, for each point to be clustered within the cluster, according to the clustering method in Steps 321 to 325, the adjacent points to be clustered owned by each point to be clustered and the connection relationships between all points to be clustered can already be obtained; in a preferred embodiment, a specific implementation of performing graphic fitting on each cluster is as follows:

[0163] Step 3261. Classify directly according to the number of adjacent points to be clustered owned. Mark the points to be clustered with more than 3 adjacent points to be clustered as central points, the points to be clustered with 2 adjacent points to be clustered as connection points, and the points to be clustered with 1 adjacent point to be clustered as end points;

[0164] Step 3262. According to the connection relationships, for all central points, mark all the connection points connecting the central point to other central points or end points as one group, and perform fitting on all the points to be clustered within one group; obviously, each central point is connected to at least 3 groups of points to be clustered. If two non-coincident straight lines are fitted out among the 3 groups of points to be clustered connected by a certain central point, it indicates that there is a broken line;

[0165] For example Figure 6 shown, a specific implementation of a cluster after clustering is given. Figure 6 In it, 3A and 3B respectively represent two different central points, and A1 to A5 respectively represent groups composed of different points to be clustered; Figure 6 There are two central points, 3A and 3B. The central point 3A is connected to three groups of points to be clustered, A1, A2, and A3, and the central point 3B is connected to three groups of points to be clustered, A3, A4, and A5. Each group of points to be clustered includes multiple connection points 2 and 1 end point 1. After fitting, it is found that two non-coincident straight lines are fitted out for groups A1 and A2, indicating that there is a broken line within this cluster.

[0166] Step 3263. Prioritize the recognition of the clusters with broken lines, that is, recognize the cluster where the central point is located; if there is no broken line in each cluster, directly recognize each cluster in sequence.

[0167] Since the broken line is a typical outer contour feature unique to artificial objects, almost all high-speed flying aircraft other than balloon-like ones and some balloon-like aircraft in modern aviation technology have broken lines. For some balloon-like aircraft without broken line contours, their movement speed is slow and the requirement for recognition speed is not high, so other methods can be used for recognition. Therefore, through the above method, the present invention can quickly identify the broken line, and then deeply identify the cluster where the central point where the broken line is located, which can greatly reduce the recognition area in the picture and improve the recognition efficiency. Deep recognition can adopt artificial intelligence recognition technology in the prior art, and refine the recognition of small pictures where each cluster is located by training a dedicated image recognition model, such as convolutional neural network (CNN), support vector machine (SVM), multi-step phase shift method, etc.

[0168] In a preferred embodiment, the joint of gray-scale regions can be further performed using the endpoints of the wavebands to be screened obtained in step 3;

[0169] Specifically:

[0170] For each row, each segment separated by the wavebands to be screened is marked as a merged segment, and the average value of the pixels in each merged segment is calculated as the pixel mean of the merged segment;

[0171] Set the second adjacent difference threshold C2. Among adjacent rows, the merged segments with overlapping column coordinates are used as inter-row adjacent merged segments. If the difference between the pixel means of the merged segments between two different inter-row adjacent merged segments is less than the second adjacent difference threshold C2, then the clustering points corresponding to the wavebands to be screened on both sides of these two inter-row adjacent merged segments are taken as the same cluster, and those greater than the second adjacent difference threshold are not taken as the same cluster.

[0172] As Figure 7 shown, a partial structure of four rows of digital vision waves is given. The first row includes two wavebands to be screened SC and two merged segments H5 and H1. The numbers in parentheses represent the pixel means of the merged segments, which are 67 and 68 respectively. The second row also has a merged segment H2 with a pixel mean of 69. Set the second adjacent difference threshold C2 = 3. Then the differences between the pixel means of the merged segments H1, H5 and the adjacent merged segment H2 in the second row are all less than the second adjacent difference threshold C2. However, only the column coordinates of the merged segments H1 and H2 have overlapping parts, while the column coordinates of the merged segment H5 and the merged segment H2 do not have overlapping parts. Therefore, the clustering points corresponding to the wavebands to be screened on both sides of the merged segments H1 and H2, that is, the endpoints or midpoints of these wavebands to be screened, are clustered in the same cluster. Since the merged segment H5 does not overlap with the merged segment H2 in the vertical coordinate and is not continuous with the merged segment H1 in the same row, although the pixels are close, it also indicates that this merged segment is not a continuously existing pattern part in terms of visual effect with the same row and adjacent rows. Therefore, the merged segment H5 does not form a continuous line or pattern with the merged segments H1 and H2, and shape fitting is not performed.

[0173] And so on, the merged segment H1 in the first row, the merged segment H2 in the second row, the merged segment H3 in the third row, and the merged segment H4 in the fourth row all have overlapping column coordinates and the difference in pixel mean values is less than the second adjacent difference threshold C2, indicating that these merged segments have similar pixels and adjacent positions, belonging to the same pattern. Therefore, the clustering points corresponding to the bands to be screened on both sides of these merged segments are classified into one cluster.

[0174] There is also a merged segment H6 in the third row with a pixel mean value of 28, and a merged segment H7 in the fourth row with a pixel mean value of 44. Although there is an overlapping part in their column coordinates, the difference in their pixel mean values is greater than the second adjacent difference threshold C2 = 3, indicating that the pixel difference is large and they probably do not belong to the same pattern. For example, they may be an aircraft and a cloud. Therefore, the clustering points corresponding to the bands to be screened on both sides are not initially classified into the same cluster.

[0175] Through the combination of gray-scale regions, the interference between different pixel boundaries can be avoided, and the boundary points of the regions with similar pixels, that is, the clustering points corresponding to the bands to be screened, are classified into the same cluster. For example, Figure 9 The aircraft marked by point F in the upper part is located on the boundary of the cloud, causing the contour line of the aircraft and the boundary line of the cloud to be clustered during clustering, making it difficult for subsequent automatic image recognition.

[0176] The combination of gray-scale regions utilizes the characteristic that there is an obvious difference between the average pixels of the cloud and the average pixels of the aircraft. After the combination of gray-scale regions, the clustering points corresponding to the boundary contours on both sides of the regions with similar pixels between adjacent rows are merged into the same cluster, and the clustering points corresponding to the regions with a large pixel difference are removed, only retaining the clustering points corresponding to the regions with similar pixels, making it easier to quickly identify the aircraft.

[0177] As Figure 9 shown, a schematic diagram of a specific implementation manner of the image recognition method described in the present invention is given. Figure 9 The upper part is the original image, and the lower part is the image obtained after recognition processing. Due to the use of the broken line and gray level screening based on clustering, the bands to be screened caused by the cloud layer are removed, and the aircraft located on the cloud background but having broken lines and gray differences can be quickly identified.

[0178] In another preferred implementation manner, for the clusters after clustering, especially for the clusters where broken lines are identified, the positions of all clustering points are judged, and the row and column numbers where each clustering point is located are used for judgment to determine whether the cluster includes clustering points located at the image boundary. If not, the cluster is preferentially deeply recognized.

[0179] Since clusters including broken lines may be generated by ground objects, such as due to the angle of the camera, for ground high-rise buildings such as skyscrapers and power transmission towers, and ground artifacts near the visual field boundary, these objects will also be clustered into clusters with broken line boundaries. However, since these objects are not aircraft suspended in the air, the clusters will all extend to the boundary of the image. By screening out clusters without clustering points at the image boundary, the interference of these ground artifacts can be excluded, effectively reducing the misidentification of ground artificial objects and improving the recognition speed. Even if the aircraft may be located at the image boundary, for fast-flying aircraft, it is very likely to appear at non-boundary positions in the image. For each frame of the image, even if the situation where the aircraft is still located at the image edge may be screened out, the aircraft will still be found in the subsequent frames or previous frames at the center of the image. By screening out clusters without clustering points at the image boundary in these frames and then performing recognition, the aircraft can still be recognized while accelerating the recognition speed.

[0180] By adopting the visual image processing method based on brain-like processing described in the present invention, endpoint extraction and clustering are performed by directly identifying the boundary of pixel point difference changes, and the clusters after clustering are recognized. The overall recognition of a large image is transformed into the recognition of small contour lines, which improves the image recognition speed and reduces the hardware requirements for image recognition. It is suitable for the recognition of aircraft in the sky background with relatively pure background and high requirements for recognition speed.

[0181] Those skilled in the art should understand that the embodiments of the present invention can be provided as a method, a system, or a computer program product. Therefore, the present invention can take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware aspects. Moreover, the present invention can take the form of a computer program product implemented 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.

[0182] The present invention is described with reference to the flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to the embodiments of the present invention. It should be understood that each process and / or block in the flowchart and / or block diagram can be implemented by computer program instructions, and the combination of processes and / or blocks in the flowchart and / or block diagram can also be implemented by computer program instructions. These computer program instructions can be provided to the processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing devices to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing devices generate means for realizing the functions specified in Figure 1 one process or multiple processes and / or blocks Figure 1 one block or multiple blocks.

[0183] The foregoing are various preferred embodiments of the present invention. If the preferred implementation manners in each preferred embodiment are not obviously self-contradictory or premised on a certain preferred implementation manner, each preferred implementation manner can be arbitrarily superimposed and combined for use. The said embodiments and the specific parameters in the embodiments are only for clearly expressing the inventor's invention verification process, and are not intended to limit the patent protection scope of the present invention. The patent protection scope of the present invention still takes its claims as the criterion. All equivalent structural changes made by using the content of the specification and drawings of the present invention should, by the same token, be included in the protection scope of the present invention.

Claims

1. A visual image processing method based on brain-like processing, characterized in that, It includes the following steps: Step 1. Divide the image into multiple rows and columns by pixels, and scan each row row by row. Taking the pixel value of each point in the row as the ordinate and the column number where the point is located as the abscissa, obtain the digital vision wave of this row; Step 2. Set the first merging threshold N1 and the first adjacent difference threshold C1; screen out the bands to be screened in each row; The method for judging that each band in the digital vision wave is marked as a non-screening band is: the difference between the abscissas of all points in the band and the starting point of the band does not reach the first merging threshold N1, and the difference between the maximum and minimum values of the ordinates of all points in the band does not reach the first adjacent difference threshold C1; The bands composed of the remaining points are marked as bands to be screened; Scan the picture row by row to obtain the bands to be screened in each row, and scan column by column to obtain the bands to be screened in each column; Step 3. Perform regional fitting on the bands to be screened and output the recognition image. Specifically: Step 31. Take the center points, as well as the front and rear endpoints, of each band to be screened, and set the second merging threshold N2. If the difference between the front and rear endpoints is less than the second merging threshold N2, represent the band to be screened directly with the center point; If the difference between the front and rear endpoints is not less than the second merging threshold N2, represent the band to be screened with the front and rear endpoints; Step 32. Perform graphic fitting recognition on the center points and endpoints of the bands to be screened retained in each row and column to obtain the image.

2. The visual image processing method based on brain-like processing according to claim 1, wherein The specific method for obtaining the bands to be screened in Step 2 is: Step 21. Read the pixel values of the first point and the second point in the digital vision wave, and set the smaller one of the pixel values of the two points as the minimum value and the other as the maximum value; Step 22. Compare with the minimum value and the maximum value respectively from the third point to the N1th point; if it is less than the minimum value, update the minimum value with the pixel value of this point; if it is greater than the maximum value, update the maximum value with the pixel value of this point; Do not update in other cases; When Step 22 reaches the N1 + 1th point, enter Step 23; Step 23. Start from the jth point, j = N1 + 1, N1 + 2…; Define the maximum value and the minimum value of the jth point as the maximum value and the minimum value respectively among all the ordinates of the N1 points from the j - N1th point to the jth point; Calculate the difference between the maximum value and the minimum value of the jth point. If it is greater than the first adjacent difference threshold C1, start marking as a band to be screened from the jth point and continue to the next point, that is, the j + 1th point; until the absolute value of the difference between the maximum value and the minimum value of a certain point is not greater than the first adjacent difference threshold C1, the band to be screened ends; Step 24. Mark the remaining points that are not bands to be screened as non-screening bands.

3. The visual image processing method based on brain-like processing according to claim 2, wherein In Step 2, after Step 24, it further includes backtracking screening. Specifically: Starting from the second row, compare the column numbers (i.e., ordinates) of the bands to be screened in the Kth row and the row above it, that is, the K - 1th row, where K is greater than or equal to 2; Step 251. First, judge whether there is partial overlap between a certain band to be screened SCi in the K - 1th row and a certain band to be screened in the Kth row. If so, retain the band to be screened in the K - 1th row; at the same time, mark the band to be screened SCj1 that overlaps with the band to be screened SCi in the Kth row and the K - 1th row as a screened band to be screened; Otherwise, enter Step 252; Step 252. Continue to check the difference in the ordinate values of the two endpoints of the to-be-screened band SCi in the (K - 1)-th row and the endpoints of the nearest to-be-screened band in the K-th row. When the difference in ordinate values does not exceed the set column threshold LT, retain the to-be-screened band SCi in the (K - 1)-th row, and mark the to-be-screened band SCj2 in the K-th row that is closest to the to-be-screened band SCi as the screened to-be-screened band; when the difference in ordinate values exceeds the set column threshold LT, re-mark the to-be-screened band in the (K - 1)-th row as a non-screening band and prevent it from entering the subsequent steps. Step 253. Traverse all the to-be-screened bands in the (K - 1)-th row. Step 254. Starting from the third row, repeat the above Steps 251 to 253 for the remaining to-be-screened bands in the previous row except for the screened to-be-screened bands.

4. The visual image processing method based on brain-like processing according to claim 2, wherein In Step 2, after Step 24, it further includes backtracking screening, specifically: Starting from the K-th row, where K is greater than or equal to M, and M is the defined minimum row span. Define the adjacent row detection process, including the following steps: Step 251. First, determine whether there is partial overlap between a certain to-be-screened band SCi in the (K - 1)-th row and a certain to-be-screened band in the K-th row. If so, retain the to-be-screened band SCi in the (K - 1)-th row; at the same time, mark the to-be-screened band SCj1 in the K-th row that overlaps with the to-be-screened band SCi in the (K - 1)-th row as the screened to-be-screened band. Otherwise, proceed to Step 252. Step 252. Continue to check the difference in the ordinate values of the two endpoints of the to-be-screened band SCi in the (K - 1)-th row and the endpoints of the nearest to-be-screened band in the K-th row. When the difference in ordinate values does not exceed the set column threshold LT, retain the to-be-screened band SCi in the (K - 1)-th row, and mark the to-be-screened band SCj2 in the K-th row that is closest to the to-be-screened band SCi as the screened to-be-screened band; when the difference in ordinate values exceeds the set column threshold LT, re-mark the to-be-screened band in the (K - 1)-th row as a non-screening band and prevent it from entering the subsequent steps. Step 253. Traverse all the to-be-screened bands in the (K - 1)-th row. The backtracking screening for a certain to-be-screened band SCi1 in the K-th row specifically includes the following steps: Start with a certain to-be-screened band SCi1 in the (K - 1)-th row; perform the adjacent row detection process described in Steps 251 to 253. If no screened to-be-screened band can be obtained in the K-th row, re-mark the to-be-screened band SCi1 in the (K - 1)-th row as a non-screening band and terminate the backtracking screening for the to-be-screened band SCi1. If a screened to-be-screened band is obtained in the K-th row, mark the to-be-screened band SCi1 in the (K - 1)-th row as the backtracking target. For all the to-be-screened bands in the (K - 2)-th row, continue with the following Step 255. Step 255 is specifically: perform the adjacent row detection process described in Steps 251 to 253, and determine whether there is a certain to-be-screened band in this row, i.e., the (K - 2)-th row, whose corresponding screened to-be-screened band is the backtracking target. If not, re-mark the to-be-screened band SCi1 in the K-th row as a non-screening band. If there is such a to-be-screened band, mark this to-be-screened band SCi2 in this row, i.e., the (K - 2)-th row, as the new backtracking target. Repeat step 255 line by line for lines from the (K - 3)-th line to the (K - M + 1)-th line; determine whether there is a to-be-screened band in this line whose corresponding screened to-be-screened band is the backtracking target. If there is no such band in any line, re-mark the to-be-screened band SCi1 in the K-th line as a non-screening band and terminate the backtracking screening. If there is such a to-be-screened band, mark this to-be-screened band in this line as the new backtracking target and go to the previous line to perform step 255 again until the (K - M + 1)-th line.

5. The visual image processing method based on brain-like processing according to claim 1, wherein Specifically, step 32 is as follows: Step 321. Take the center points, as well as the front and back end points of each to-be-screened band as the to-be-clustered point library. Step 322. Arbitrarily select a to-be-clustered point in the to-be-clustered point library as the target point, and find all adjacent to-be-clustered points of this clustered point in the to-be-clustered point library. The to-be-clustered point library is the point library composed of the center points and end points of the reserved to-be-screened bands. The adjacent to-be-clustered points are points whose spatial distance from the target point is less than the spatial threshold. Step 323. Remove the selected adjacent to-be-clustered points from the to-be-clustered point library; for each selected adjacent to-be-clustered point, take it as the target point and continue to find all adjacent to-be-clustered points of this target point in the to-be-clustered point library. Step 324. Repeat step 323 until no adjacent to-be-clustered points can be found in the to-be-clustered point library, and one clustering is completed. Step 325. Repeat steps 322 to 324 until all points in the to-be-clustered point library are clustered. Each clustering results in a cluster composed of multiple to-be-clustered points. Step 326. Perform the graphic fitting recognition on the to-be-clustered points within each cluster.

6. The visual image processing method based on brain-like processing according to claim 5, characterized in that The spatial distance L = [(h1 - h0) 2 + (l1 - l0) 2 1 / 2 ;​ where the spatial coordinates of the target point are (h0, l0), and h0 and l0 are the row and column numbers of the target point respectively; the spatial coordinates of the to-be-clustered point are (h1, l1), and h1 and l1 are the row and column numbers of this to-be-clustered point respectively.

7. The visual image processing method based on brain-like processing according to claim 5, wherein Specifically, step 326 is as follows: Step 3261. Directly classify according to the number of adjacent to-be-clustered points owned. Mark the to-be-clustered points with more than 3 adjacent to-be-clustered points as central points, the to-be-clustered points with 2 adjacent to-be-clustered points as connection points, and the to-be-clustered points with 1 adjacent to-be-clustered point as end points. Step 3262. According to the connection relationship, for all central points, mark all the connection points connecting this central point to other central points or end points as 1 group, and perform fitting on all the to-be-clustered points within 1 group; if among the 3 groups of to-be-clustered points connected by a certain central point, there are two groups that fit out non-coincident straight lines, it indicates that there is a broken line. Step 3263. Prioritize identifying the cluster with a broken line, that is, identify the cluster where this central point is located; if there is no broken line in each cluster, directly identify each cluster in sequence.

8. The visual image processing method based on brain-like processing according to claim 5, wherein After all clusters are obtained in step 325, it further includes the gray-level joint correction for the clusters, specifically: For each line, mark each segment separated by the to-be-screened difference band as a merged segment, and calculate the average value of the pixels within each merged segment as the pixel mean of the merged segment. Set the second adjacent difference threshold C2. Among adjacent rows, the merged segments with overlapping column coordinates are used as the inter-row adjacent merged segments. If the difference in the mean values of the pixels in the merged segments between two inter-row adjacent merged segments in different rows is less than the second adjacent difference threshold C2, then the clustering points corresponding to the to-be-screened difference bands on both sides of these two inter-row adjacent merged segments are taken as the same cluster, and if it is greater than the second adjacent difference threshold, they are not taken as the same cluster.

9. The visual image processing method based on brain-like processing according to claim 5, wherein In step 326, for all the clusters after clustering, traverse each cluster and judge the positions of all the clustering points within each cluster to determine whether there are clustering points located on the image boundary. If not, then give priority to deep recognition of this cluster.

10. The visual image processing method based on brain-like processing according to claim 1, wherein In step 2, merge the non-screening bands, and for all the points in the merged band, the ordinate is taken as the average value of the ordinates of all the points in the band.

Images