Liquid crystal display dead pixel detection method and system

Abstract

The invention discloses a liquid crystal display dead pixel detection method and system, and relates to liquid crystal display detection, and the method comprises the steps: obtaining a dead pixel position region of a liquid crystal display according to a preset test scheme; constructing a physical topology model of the liquid crystal screen according to the physical parameters of the liquid crystal screen; segmenting the dead pixel position area by adopting an improved watershed algorithm to obtain a segmentation result; performing feature analysis on the segmentation result by adopting a pre-constructed defective pixel feature knowledge base, determining a defective pixel name and a defective pixel inducement, and obtaining a detection result; a physical topology model is utilized, and a flow rule based on a physical connection path is adopted, so that the watershed algorithm only diffuses along the physical connection path, and a segmentation result is obtained; aiming at the problem of low dead pixel detection precision caused by the over-segmentation phenomenon easily generated by a watershed algorithm in the prior art, physical constraints are established by using a regular structure of an LCD pixel array to improve the watershed algorithm, so that the dead pixel detection precision is improved.

Description

Technical Field

[0001] The present application relates to the field of liquid crystal screen detection, and in particular to a method and system for detecting bad pixels on a liquid crystal display screen. Background Art

[0002] In the field of LCD screen bad pixel detection technology, traditional methods mainly rely on manual visual inspection or simple image processing algorithms, such as threshold segmentation, edge detection, etc. With the development of computer vision technology, more complex image processing algorithms have been introduced into the field of bad pixel detection, among which the watershed algorithm is widely used in the division and identification of bad pixel areas due to its good image segmentation ability.

[0003] The watershed algorithm is based on the concept of watershed in geography. It regards the image as a topographic map. The areas with high grayscale values ​​are like peaks, and the areas with low grayscale values ​​are like valleys. The algorithm simulates the process of water gradually rising from a low point. When the water in different valleys is about to connect, a watershed line is established to achieve regional segmentation of the image. This method can effectively segment different areas under ideal conditions and is suitable for determining the boundaries between bad pixels and normal display areas.

[0004] However, the traditional watershed algorithm has a serious over-segmentation defect when processing LCD screen bad pixel images. Since the algorithm is completely based on the grayscale gradient change of the image and ignores the physical structural characteristics of the LCD screen, the bad pixel that should be a single fault area is mistakenly segmented into multiple unrelated sub-areas. In particular, when the bad pixel spreads along the drive line or presents an irregular shape, the traditional watershed algorithm will excessively establish watershed lines at the gradient change, resulting in a large number of meaningless small areas, which seriously affects the subsequent bad pixel classification and cause analysis.

[0005] For example, the relevant patent document CN119068787A discloses a method, device and storage medium for detecting bad pixels on a liquid crystal screen. Based on a preset test scheme, the test equipment is controlled to test the target liquid crystal screen, and the screen characteristic image of the target liquid crystal screen is obtained. The actual gradient value matrix of the target liquid crystal screen is constructed according to the screen characteristic image; the bad pixel position of the target liquid crystal screen is analyzed according to the actual gradient value matrix of the target liquid crystal screen and the standard gradient value matrix, and the bad pixel position area in the screen characteristic image is obtained; the screen characteristic image is segmented to obtain the sub-characteristic image corresponding to the bad pixel position area in the target liquid crystal screen; the feature analysis of each sub-characteristic image is performed based on the bad pixel feature knowledge tree graph network, and the bad pixel name and bad pixel inducement of all bad pixel position areas in the target liquid crystal screen are obtained. However, this scheme adopts the traditional watershed algorithm, which may mistakenly segment a single bad pixel area into multiple sub-areas, so the detection accuracy of this scheme needs to be further improved. Summary of the invention

[0006] In view of the fact that the watershed algorithm in the prior art is prone to over-segmentation, resulting in low bad pixel detection accuracy, the present application provides a method and system for bad pixel detection on a liquid crystal display screen. By utilizing the regular structure of the LCD pixel array, physical constraints are established to improve the watershed algorithm, thereby improving the bad pixel detection accuracy.

[0007] The purpose of this application is achieved through the following technical solutions.

[0008] One aspect of the present application provides a method for detecting bad pixels on a liquid crystal display screen, comprising: S1, obtaining a bad pixel location area of ​​a liquid crystal screen according to a preset test scheme; S2, constructing a physical topology model of the liquid crystal screen according to physical parameters of the liquid crystal screen; S3, segmenting the bad pixel location area using an improved watershed algorithm to obtain a segmentation result; S4, performing feature analysis on the segmentation result using a pre-constructed bad pixel feature knowledge base to determine the bad pixel name and the bad pixel inducement to obtain a detection result; wherein, segmenting the bad pixel location area using the improved watershed algorithm comprises: using a physical topology model and adopting a flow rule based on a physical connection path so that the watershed algorithm diffuses only along the physical connection path to obtain a segmentation result;

[0009] Furthermore, the physical parameters include RGB pixel arrangement, sub-pixel structure and drive circuit layout; the bad pixel feature knowledge base includes feature parameters, cause analysis and identification rules of various types of bad pixels;

[0010] Further, S1, obtaining the bad pixel location area of ​​the LCD screen according to a preset test scheme, including: determining the standard gradient value corresponding to each pixel in the LCD screen according to the preset test scheme, and generating a standard gradient value matrix; the standard gradient value represents the grayscale gradient value corresponding to the pixel under preset normal working conditions; obtaining a test image of the LCD screen under the test conditions; the test image represents the display image of the LCD screen in the test mode; constructing an actual gradient value matrix according to the test image; and determining the bad pixel location area of ​​the LCD screen by comparing the standard gradient value matrix and the actual gradient value matrix.

[0011] Further, S2, constructing a physical topology model of the LCD screen according to the physical parameters of the LCD screen, including: obtaining the RGB pixel arrangement of the LCD screen, determining the physical connection relationship between adjacent pixels, and the positional relationship between adjacent pixels; establishing a two-dimensional pixel coordinate system according to the physical connection relationship between adjacent pixels and the positional relationship between adjacent pixels; determining the physical connection path at the sub-pixel level according to the physical boundaries of the LCD sub-pixels; obtaining the drive circuit layout data of the LCD screen, determining the connection relationship between the pixel points and the gate lines and the data lines according to the physical routing structure of the drive circuit, and identifying the pixel point groups sharing the same drive circuit by tracing the drive signal transmission path; constructing a pixel connection weight matrix according to the RGB pixel arrangement, sub-pixel and drive circuit layout data; generating a gradient weight matrix containing physical constraints according to the pixel connection weight matrix and the standard gradient value matrix; establishing a drive line fault model according to the drive circuit layout data and the transmission characteristics and attenuation law of the drive signal on the line; constructing a physical topology model of the LCD screen according to the two-dimensional pixel coordinate system, the physical connection path at the sub-pixel level, the pixel connection weight matrix and the drive line fault model.

[0012] Among them, sub-pixel is the basic display unit that constitutes a single complete pixel in the liquid crystal display. It is usually composed of red (R), green (G), and blue (B) light-emitting units. Each sub-pixel can independently control its brightness level. Sub-pixel is the smallest physical unit for building a physical topological model. Its physical boundary and arrangement structure determine the basic physical constraints of signal transmission and fault propagation, and are the basis for accurately analyzing the formation and diffusion mechanism of bad pixels.

[0013] The physical connection path refers to the signal transmission channel or mutual influence channel on the physical level between adjacent pixels or sub-pixels in the LCD screen, which represents the physical path through which the fault signal between display units may propagate. The physical connection path is determined based on the actual physical structure of the LCD screen, including the wire connection of the TFT array layer, the stress conduction between liquid crystal molecules, the electric field coupling of the shared electrode, etc. The physical connection path defines the legal flow channel of the "water flow" in the watershed algorithm, ensuring that the algorithm segmentation results meet the actual physical structure constraints and preventing the wrong segmentation of areas that should be physically connected.

[0014] The connection relationship between the pixel point and the gate line and data line describes the physical and electrical connection between the display unit and the drive signal transmission line, reflecting the addressing and driving mechanism of the TFT liquid crystal screen. The connection relationship between the pixel point and the drive line reveals the physical channel of signal transmission in the liquid crystal screen. It is the key basis for identifying bad points of drive line faults and predicting the fault propagation path. It is also an important reference for constructing the pixel connection weight matrix.

[0015] A pixel group sharing the same drive line refers to a set of pixel points controlled by the same gate line or data line. These pixel points are physically related in the drive signal transmission path and may exhibit similar fault characteristics.

[0016] Among them, the gradient weight matrix containing physical constraints generated according to the pixel connection weight matrix and the standard gradient value matrix provides an accurate gradient value basis for the watershed algorithm, making the path selection of the immersion process more in line with the physical characteristics of the LCD screen. This gradient control mechanism under physical constraints fundamentally ensures that the water flow only diffuses along the physically connected path, avoiding the erroneous diffusion across physically isolated areas in traditional algorithms.

[0017] Furthermore, a pixel connection weight matrix is ​​constructed, including: obtaining the physical distance between pixels and the sharing of driving lines; setting the connection weight to N1 for pixels whose physical distance is less than a threshold and share driving lines; setting the connection weight to N2 for pixels whose physical distance is less than a threshold but do not share driving lines; and setting the connection weight to N3 for pixels whose physical distance exceeds the threshold.

[0018] Furthermore, N1 is greater than N2, and N2 is greater than N3.

[0019] Further, S3, an improved watershed algorithm is used to segment the bad point location area to obtain a segmentation result, including: using the bad point location area determined in S1 as the input area of ​​the watershed algorithm; using the physical connection path in the physical topology model constructed in S2 as the flow constraint condition of the watershed algorithm; setting the gradient value of the watershed algorithm according to the gradient weight matrix containing physical constraints generated in S2; setting the initial water level line, and using the pixel point with the lowest gradient value as the initial immersion point; during the immersion process, controlling the water flow diffusion path according to the pixel connection weight matrix, setting the diffusion priority of the path with a connection weight of N1 to P1, and the diffusion priority of the path with a connection weight of N2 to P1. Set it to P2, and set the diffusion priority of the path with connection weight N3 to P3; use the driving line fault model in S2 to predict the potential bad pixel diffusion direction, and adjust the regional growth strategy of the watershed algorithm; when the water level rises to the preset height, detect the minimum pixel distance between adjacent sub-regions, and establish a watershed line based on the minimum pixel distance; according to the established watershed line, for the identified pixel point groups sharing the same driving line, analyze the corresponding continuity in the two-dimensional pixel coordinate system, and assign the same regional label to the pixel groups that are continuously distributed and not separated by the watershed line; merge the pixel groups with the same regional label to form multiple independent bad pixel sub-regions as the segmentation result.

[0020] Among them, according to the established watershed line, the continuity analysis of the pixel point groups sharing the same driving line is carried out, and the pixel groups that are continuously distributed and not separated by the watershed line are given the same area mark, which ensures that each physically connected bad pixel area is correctly identified as a whole from the technical principle. The mechanism of merging pixel groups with the same area mark to form independent bad pixel sub-areas avoids the bad pixel fragmentation problem that may occur in traditional methods.

[0021] Further, P1 is greater than P2, and P2 is greater than P3;

[0022] Furthermore, when the water level rises to a preset height, the minimum pixel distance between adjacent sub-areas is detected, and a watershed line is established based on the minimum pixel distance, including: calculating the minimum pixel distance between each sub-area formed during the immersion process; when the minimum pixel distance between adjacent sub-areas is less than a threshold, obtaining the position information of the pixel points of the adjacent sub-areas in the two-dimensional pixel coordinate system constructed in S2; according to the position information, obtaining the connection relationship between the pixel points and the gate lines and data lines in the driving circuit layout data in S2 of the pixel points in the adjacent sub-areas; according to the position information, and the connection relationship between the pixel points and the gate lines and data lines, obtaining the physical connection path determined in S2 for the pixel points in the adjacent sub-areas, and the physical connection path is used to determine whether there is a direct physical path between the adjacent sub-areas; when the adjacent sub-areas share the same driving circuit and there is a direct physical path on the physical connection path, the watershed line is not established; when the adjacent sub-areas do not share the same driving circuit or there is no direct physical path on the physical connection path, the watershed line is established.

[0023] Among them, the minimum pixel distance refers to the shortest spatial distance between the boundary pixels of two adjacent sub-regions formed during the immersion process of the watershed algorithm. The calculation method is the minimum Euclidean distance between the boundary pixel sets of two sub-regions: , and All are boundary images; the distance can be calculated in pixel coordinate space or converted into physical distance (mm). The minimum pixel distance is used as a conditional parameter to trigger physical connection analysis. When the distance is less than the preset threshold, the algorithm will further analyze the physical connection characteristics of the adjacent area. This two-stage strategy significantly improves the efficiency of the algorithm.

[0024] The physical connection path refers to the signal transmission channel or fault propagation channel that may exist between pixels based on the actual physical structure of the LCD screen, reflecting the actual connection relationship between the components inside the LCD screen. It is jointly determined by the TFT array layer structure, drive line layout and sub-pixel arrangement of the LCD screen. It includes many types: electrical connection path: such as the connection between pixels sharing the same gate line or data line; molecular mechanics connection path: the connection formed by the stress transfer between liquid crystal molecules; optical connection path: the influence propagation path formed by the optical crosstalk between pixels; it has directionality and strength properties, reflecting the difficulty of physical influences propagating in different directions; it is usually represented by a weighted directed graph structure, with nodes as pixels, edges as physical connections, and weights representing connection strength; it is pre-calculated and stored in the S2 stage based on the sub-pixel physical boundaries and drive line layout data. The physical connection path provides strict physical structure constraints to ensure that the segmentation results of the watershed algorithm conform to the actual physical connection characteristics of the LCD screen and avoid incorrect segmentation of areas that should be physically connected.

[0025] A direct physical path refers to an unobstructed physical connection channel between adjacent sub-regions, which allows the physical signal or fault impact related to the bad pixel to be directly transmitted from one region to another, indicating that these sub-regions belong to the same bad pixel region in physical essence. The judgment criteria include: the boundary pixels in adjacent sub-regions share the same drive line (gate line or data line); there is a continuous physical connection path between sub-regions, and there is no physical isolation structure blocking; the weight value of the connection path is higher than the preset threshold, indicating that the connection strength is sufficient to support fault propagation; the verification method usually uses a path tracing algorithm, such as A* or Dijkstra algorithm, to find the shortest path between two regions on the physical connection path map. The existence of a direct physical path is the key condition for determining whether to merge adjacent sub-regions, ensuring that the segmentation result of the watershed algorithm conforms to the physical formation mechanism and propagation law of bad pixels, and preventing over-segmentation problems.

[0026] The watershed line is the regional boundary line formed in the improved watershed algorithm, which is used to separate different bad pixel sub-regions. It is established based on a comprehensive evaluation of image gradient, physical distance and physical connectivity, rather than relying solely on image gradient. The difference from the traditional watershed line: the traditional watershed line is formed only based on image gradient, which is easy to cause over-segmentation; the improved watershed line comprehensively considers the physical connection relationship and has physical significance; the minimum pixel distance between adjacent regions is less than the threshold (preliminary condition); adjacent regions do not share the same driving line or there is no direct physical path (decisive condition); the precise positioning of the watershed line usually adopts the middle position of the boundary of the adjacent sub-region, or is determined based on the local gradient maximum; each watershed line has a weight attribute, which reflects the confidence of the segmentation decision, and the high-weight watershed line has higher reliability. The watershed line is the final boundary of the bad pixel region segmentation, which determines the range and shape of each bad pixel sub-region in the detection result. Compared with the traditional method, the watershed line based on physical constraints more accurately reflects the real physical boundary of the bad pixel, significantly improving the accuracy of bad pixel segmentation and the reliability of subsequent analysis.

[0027] Another aspect of the present application also provides a liquid crystal display screen bad pixel detection system, which is used to execute a liquid crystal display screen bad pixel detection method of the present application.

[0028] Compared with the prior art, the advantages of this application are:

[0029] (1) The traditional watershed algorithm has a serious over-segmentation defect when processing images of bad pixels on LCD screens. Since it is based only on the grayscale gradient changes of the image and completely ignores the physical structural characteristics of the LCD screen, the bad pixels that should be a single fault area are incorrectly segmented into multiple unrelated sub-areas. This application constructs a physical topology model to convert the RGB pixel arrangement, sub-pixel structure and drive line layout of the LCD screen into algorithm constraints, so that the flow rules of the watershed algorithm are consistent with the physical connection characteristics of the LCD screen, and avoids the over-segmentation phenomenon caused by the traditional watershed algorithm that does not consider the physical structure from a technical principle.

[0030] During the immersion process of the watershed algorithm, the water diffusion priority (P1, P2, P3) is controlled by the pixel connection weight matrix (N1, N2, N3), so that the pixels of the shared driving line are merged first, fundamentally preventing the bad pixels that should be physically classified into one category from being incorrectly segmented.

[0031] (2) The traditional method for determining the boundary of a bad pixel has the defects of blurred boundary and boundary offset. It only relies on the pixel grayscale gradient without considering the physical separation structure of the LCD panel, and cannot accurately locate the boundary of the bad pixel in the area where the gradient change is not obvious. The sub-pixel level physical connection path established by this application based on the physical parameters of the LCD screen provides a strict physical basis for the establishment of the watershed line, so that the segmentation boundary is highly consistent with the actual physical structure of the LCD screen.

[0032] When the water level rises to a preset height, it intelligently determines whether to establish a watershed line by judging whether adjacent sub-areas share a driving line and whether a physical path exists, thus avoiding the problem of setting unnecessary dividing lines in physically connected areas.

[0033] (3) Traditional bad pixel detection methods have the defect of misclassification when dealing with bad pixels caused by drive line faults. They are unable to identify the physical correlation between bad pixels propagating along the same drive line, and often misjudge multiple bad pixels that should belong to the same fault source as independent faults. This application uses the drive line fault model to predict the direction of bad pixel diffusion and adjusts the regional growth strategy of the watershed algorithm so that the algorithm can identify bad pixels propagating along the drive line and correctly classify them as a single bad pixel area rather than multiple independent bad pixels. By reducing the gradient value weight between pixels on the same drive line and increasing the diffusion priority along the drive line direction, it is technically guaranteed that the segmentation result conforms to the physical fault propagation law of the LCD display. BRIEF DESCRIPTION OF THE DRAWINGS

[0034] The present application will be further described in the form of exemplary embodiments, which will be described in detail by the accompanying drawings. These embodiments are not restrictive, and in these embodiments, the same number represents the same structure, wherein:

[0035] Figure 1 is an exemplary flow chart of a method for detecting bad pixels of a liquid crystal display screen according to some embodiments of the present application;

[0036] Figure 2 is an exemplary flow chart of obtaining a bad pixel location area according to some embodiments of the present application;

[0037] Figure 3 is an exemplary flow chart of constructing a physical topology model of a liquid crystal screen according to some embodiments of the present application;

[0038] Figure 4 This is an exemplary flowchart of obtaining segmentation results according to some embodiments of the present application. DETAILED DESCRIPTION

[0039] The method and system provided in the embodiments of the present application are described in detail below with reference to the accompanying drawings.

[0040] like Figure 1As shown, the bad pixel location area of ​​the LCD screen is obtained according to a preset test plan; the physical topology model of the LCD screen is constructed according to the physical parameters of the LCD screen; the bad pixel location area is segmented by an improved watershed algorithm to obtain a segmentation result; a pre-constructed bad pixel feature knowledge base is used to perform feature analysis on the segmentation result to determine the bad pixel name and the bad pixel cause to obtain a detection result; wherein, the bad pixel location area is segmented by an improved watershed algorithm, including: utilizing a physical topology model and adopting a flow rule based on a physical connection path so that the watershed algorithm diffuses only along the physical connection path to obtain a segmentation result.

[0041] like Figure 2 As shown, S1, according to the preset test scheme, the bad pixel location area of ​​the LCD screen is obtained, including: the preset test scheme includes a standard display pattern sequence designed for the characteristics of different types of LCD screens, usually including pure white, pure black, pure red / green / blue, grayscale gradient and checkerboard and other test patterns. For each test pattern, the grayscale gradient value that each pixel should have under each test pattern under an ideal state is determined in advance through theoretical calculation or through statistical analysis of a large number of qualified samples. The grayscale gradient value represents the rate of change of the grayscale value between adjacent pixels, and the gradient value of each pixel in the horizontal and vertical directions is usually calculated using the Sobel operator, Prewitt operator or Scharr operator. For each test pattern, a corresponding standard gradient value matrix is ​​generated, the dimension of the matrix is ​​the same as the resolution of the LCD screen, and each element in the matrix represents the standard gradient value of the pixel at the corresponding position.

[0042] Use dedicated LCD test equipment to send the same test pattern sequence as the standard test solution to the LCD screen to be tested through display interfaces such as HDMI, DVI, DisplayPort, or a driver board that directly drives the LCD screen. After each test pattern is displayed stably (usually wait for 20-50ms of stabilization time), use a high-resolution industrial camera or dedicated image acquisition equipment to capture the actual display image of the LCD screen. The captured image needs to be pre-processed, including image calibration, perspective transformation correction, noise filtering, etc., to ensure that the captured image can accurately reflect the actual display status of the LCD screen. For large-size or high-resolution LCD screens, it may be necessary to capture and stitch multiple areas to ensure that enough detail information is captured.

[0043] For each acquired test image, the same gradient calculation method as the standard gradient value matrix (such as the same Sobel operator) is used to calculate the actual gradient value of each pixel in the image. If the test image is a color image, it is first decomposed into three channels, R, G, and B, and the gradient values ​​of each channel are calculated separately. Then, according to the sub-pixel structure characteristics of the LCD screen, the gradient values ​​of the three channels are synthesized into a comprehensive gradient value. Considering the possible position offset during the camera acquisition process, the image registration technology is used to ensure that the test image pixels are strictly corresponding to the corresponding positions in the standard gradient value matrix. The constructed actual gradient value matrix also has the same dimension as the LCD screen resolution, and each element represents the actual gradient value of the pixel at the corresponding position in the test image.

[0044] For each pixel, the difference between its actual gradient value and the standard gradient value is calculated to form a difference matrix. A reasonable threshold range is set. If the difference value of a pixel exceeds the preset threshold, it is marked as a potential bad pixel. In order to avoid false positive detection caused by ambient light interference and measurement errors, a comprehensive analysis of multiple test results is usually adopted, that is, only when a pixel is abnormal under multiple test patterns, it is determined to be a bad pixel. Connectivity analysis is performed on the marked potential bad pixel areas, and adjacent bad pixel pixels are combined into bad pixel areas. Each area has clear boundary coordinates and internal pixel lists. Finally, the bad pixel location area information is output, including the location coordinates (upper left corner and lower right corner coordinates), area (number of pixels), shape features (such as rectangularity, circularity), etc. of each bad pixel area as input for subsequent analysis.

[0045] like Figure 3 As shown, S2, based on the physical parameters of the LCD screen, a physical topology model of the LCD screen is constructed; the physical parameters include RGB pixel arrangement, sub-pixel structure and drive circuit layout, including: first, the RGB pixel arrangement of the LCD screen needs to be obtained to determine the physical connection relationship between adjacent pixels: the RGB sub-pixel arrangement of the LCD screen is obtained from the LCD screen specification or through microscope measurement. Common arrangements include RGB stripe arrangement, RGB triangle arrangement, PenTile arrangement and RGBW quaternary arrangement. For the RGB stripe arrangement, adjacent pixels usually have a direct physical connection relationship in the horizontal direction, and are connected through the electrodes of the TFT array layer in the vertical direction. For the PenTile arrangement, special consideration needs to be given to the sub-pixel sharing situation. The physical connection relationship between adjacent pixels is more complicated, usually manifested as an irregular grid structure. Measure and record the physical pixel pitch (Pixel Pitch) and sub-pixel pitch. These parameters directly affect the physical distance and connection strength between adjacent pixels. At the same time, determine the positional relationship between adjacent pixels, including 8-connected or 4-connected neighborhood relationships, as well as the physical distance and actual connection strength in each direction.

[0046] According to the physical connection relationship and position relationship between adjacent pixels, a two-dimensional pixel coordinate system is established: the upper left corner pixel of the LCD screen is taken as the origin (0, 0), and a rectangular coordinate system is established. The horizontal axis represents the column position of the pixel, and the vertical axis represents the row position of the pixel. For LCD screens with conventional rectangular arrangement, the coordinate system is a regular grid; for non-rectangular arrangement (such as hexagonal arrangement), it is necessary to establish a transformation matrix to map the actual physical position to a two-dimensional coordinate system. Assign a unique coordinate identifier (x, y) to each pixel, and establish a pixel index table to facilitate fast positioning and access to any pixel. Record the list of adjacent pixels of each pixel in the coordinate system, including directly adjacent first-order neighbors and indirectly adjacent high-order neighbors, and construct a pixel adjacency relationship graph.

[0047] Determine the physical connection path at the sub-pixel level based on the physical boundaries of the LCD sub-pixels: Obtain the physical structure of the sub-pixels through a high-resolution microscope or design drawings, and measure the geometric size and shape of the sub-pixel boundaries. Analyze the physical isolation structure between sub-pixels, such as the black matrix and the edge of the color filter, which will block or weaken the optical and electrical signal propagation between adjacent sub-pixels. Based on the physical boundaries of the sub-pixels, determine the sub-pixel level connection path diagram, identify which sub-pixel boundaries allow signal transmission and which boundaries are completely isolated. For sub-pixels of different colors (R, G, B), consider their differences in physical structure and electrical characteristics, and establish connection path models respectively. Integrate the sub-pixel level connection paths to form a pixel-level physical path diagram to describe the possible propagation paths of signals and faults.

[0048] Obtain the drive circuit layout data of the LCD screen and determine the connection relationship between the pixel points and the drive circuit: extract the drive circuit layout information from the technical specification documents provided by the LCD panel manufacturer, parse the pin definition and signal layout diagram in the driver IC data sheet, and extract the number of gate lines, number of data lines, routing direction and corresponding pixel mapping relationship. Obtain the PCB design of the LCD module or scan the PCB layout through an electron microscope, identify the connection points of the flexible circuit board (FPC) connecting the driver IC to the panel, and track the connection position and arrangement order of the signal line on the FPC to the edge of the panel. Use a high-resolution infrared camera to image the panel at a specific wavelength, use optical filtering technology to enhance the visibility of the TFT layer wiring, identify the physical layout structure of the gate line and data line through image processing algorithms, construct the wiring vector diagram of the gate line and data line, and record the starting point, end point and direction of each line.

[0049] Establish a two-dimensional mapping table of (row, column) → (gate_line, data_line). For the standard matrix drive structure, the mapping usually follows the regular relationship: gate_line = row_index + gate_offset, data_line = column_index + data_offset, and record the mapping abnormal points and special areas, such as the multiplexing area and the edge compensation area. Design a sequential test mode, activate a single drive line in turn, record the set of pixel coordinates that respond when each drive line is activated, and for each pixel, establish its associated gate line ID and data line ID, and construct a two-dimensional connection relationship matrix M, where M[i, j] represents the drive line connected to the pixel point (i, j).

[0050] Based on the physical structure diagram of the TFT array layer of the LCD screen, identify the location of the TFT device at each pixel, trace the physical connection path from the TFT gate to the gate line, trace the physical connection path from the TFT source to the data line, and for the common source design, specially mark the pixel group that shares the data line. According to the design type of the LCD screen (such as a-SiTFT, LTPS, OxideTFT, etc.), select the appropriate driving structure model, establish the connection relationship calculation formula that reflects the specific technical parameters, and consider the connection variants of special driving schemes (such as in-plane gate and dual-gate design).

[0051] Based on the transmission line theory, a propagation model of the driving signal on the gate line and data line is established, the position of the signal source (driver IC output terminal) and the signal transmission direction are recorded, the transmission delay and intensity attenuation of the signal reaching each pixel are calculated, and a signal propagation feature library of typical faults (such as open circuit, short circuit, high resistance) is established. Using circuit simulation tools such as SPICE, an equivalent circuit model of the driving line is constructed, the signal transmission characteristics of the gate line and data line are simulated, the propagation characteristics of the driving signal under different load conditions are analyzed, and the key nodes and signal integrity characteristics on the signal propagation path are recorded.

[0052] Record the row scanning order and column data loading order in the driving timing, analyze the set of pixels activated simultaneously in the same clock cycle, identify timing-related pixel groups that may show similar responses to timing faults, build a timing dependency graph, and describe the timing propagation sequence of the driving signal. Measure the signal integrity parameters at different positions on the gate line and data line, record the signal rise time, fall time, overshoot, ringing and other characteristics, establish the correlation between signal integrity and pixel response, and identify signal degradation sensitive areas and potential high-incidence areas of faults.

[0053] For each gate line Li, define a pixel set G(Li) = {P(r, c) | P(r, c) is connected to the gate line Li}. Normally, G(Li) contains all the pixels in the same row of the display screen. For special layouts (such as staggered gates), G(Li) may contain partial pixels in multiple rows. The same gate line group identifier is assigned to the pixels in G(Li), and a gate line group list is constructed to record the pixel coordinate set contained in each group.

[0054] For each data line Di, define a pixel set D(Di) = {P(r, c) | P(r, c) is connected to the data line Di}. Usually, D(Di) contains all the pixels in the same column of the display screen. For a multiplexed layout, D(Di) may contain alternating column pixels or pixels in a specific pattern. The pixels in D(Di) are assigned the same data line group identifier, and a data line group list is constructed to record the pixel coordinate set contained in each group.

[0055] Based on the gate line group and data line group, more complex connection relationship groups are constructed to identify the grouping of gate driver and data driver output channels. For multiple gate lines that share the same gate driver channel, advanced gate groups are constructed. For multiple data lines that share the same data driver channel, advanced data groups are constructed. These multi-level groups are particularly important for identifying bad pixels of driver IC failure types. Identify the pixel positions corresponding to the intersections of gate lines and data lines, analyze the possible failure modes caused by the intersection of specific drive lines, construct a cross-connection matrix, and record the physical characteristics and potential risks of each intersection.

[0056] According to the RGB pixel arrangement, sub-pixel and drive line layout data, a pixel connection weight matrix is ​​constructed: the physical distance between pixels is obtained, usually using Euclidean distance or Manhattan distance, and a suitable distance metric is selected in combination with the actual physical layout characteristics. The drive line sharing between pixels is analyzed to determine whether two pixels share the same gate line or data line. A physical distance threshold is set, usually 1.5 times the pixel spacing, as a reference for judging the strength of physical connection between pixels. According to the physical distance and drive line sharing, weights are assigned to the connection between each pair of pixels: for pixels whose physical distance is less than the threshold and share the drive line, a high connection weight N1 (for example, a value of 9.0) is set; for pixels whose physical distance is less than the threshold but do not share the drive line, a medium connection weight N2 (for example, a value of 3.0) is set; for pixels whose physical distance exceeds the threshold, a low connection weight N3 (for example, a value of 0.5) is set; the weight values ​​satisfy the relationship of N1>N2>N3, usually N1 is about 3 times N2, and N2 is about 6 times N3. This setting ensures that pixels that share the drive line have a significantly higher weight advantage than other connections. The connection weights between all pixel pairs are organized into a matrix form to form a pixel connection weight matrix.

[0057] According to the pixel connection weight matrix and the standard gradient value matrix, a gradient weight matrix containing physical constraints is generated: the standard gradient value matrix generated in S1 is fused with the pixel connection weight matrix to obtain the gradient value that takes into account the physical connection characteristics. For pixel pairs with high connection weights (such as adjacent pixels that share a driving line), their gradient values ​​are reduced so that they are not easy to form segmentation boundaries in the watershed algorithm. For pixel pairs with low connection weights (such as physically isolated pixels), their gradient values ​​are maintained or increased so that they are easy to form segmentation boundaries in the watershed algorithm. The specific fusion method uses the gradient value adjustment formula: adjusted gradient value = original gradient value × (1-α × connection weight), where α is the adjustment factor, usually 0.1-0.3. The generated physical constraint gradient weight matrix maintains the same size as the original image, but each element value has been fused with physical connection constraint information.

[0058] According to the drive line layout data, establish a drive line fault model: analyze the electrical characteristics and optical manifestations of different types of drive line faults, such as gate line open circuit, data line short circuit, etc. Study the transmission characteristics of the drive signal on the line, including signal delay, attenuation and crosstalk. Establish a mathematical model to describe how the fault signal propagates along the drive line, including propagation speed, attenuation coefficient and impact range. For common fault types, such as open circuit, short circuit, high resistance, leakage, etc., establish prediction models to infer the fault source and possible diffusion direction. Use historical fault data to verify and optimize model parameters to improve the accuracy of fault propagation prediction.

[0059] Finally, the above components are combined to build a complete physical topology model of the LCD screen: integrating the two-dimensional pixel coordinate system, sub-pixel-level physical connection path, pixel connection weight matrix and drive line fault model to form a multi-level physical topology model. The model is represented by a graph structure, with nodes as pixels, edges as physical connection relationships, and edge weights representing connection strength. A drive line layer is added to the model to record the subordinate relationship between each pixel and the drive line, as well as the cross-connections between drive lines. A mapping relationship from sub-pixel to pixel is established to achieve model analysis and operation at different levels of accuracy. The final physical topology model can accurately express the physical structural characteristics of the LCD screen, providing an important basis for the subsequent watershed algorithm based on physical constraints.

[0060] like Figure 4As shown, S3 uses an improved watershed algorithm to segment the bad pixel location area to obtain the segmentation result; first, the information obtained in the previous step is used as the input of the watershed algorithm: the bad pixel location area determined in S1 is used as the input area of ​​the watershed algorithm. These areas are usually a set of pixels with obvious grayscale anomalies, which have been preliminarily determined as areas where bad pixels may exist. The input area is usually represented as a binary mask image, where the pixel value of the bad pixel area is 1 and the pixel value of the non-bad pixel area is 0. In order to improve computational efficiency, each independent bad pixel location area is usually cut into sub-areas so that the algorithm can process multiple independent areas in parallel. Each sub-area will be appropriately expanded before processing to include a small number of normal pixels around it to ensure that the segmentation boundary can be accurately determined.

[0061] The physical connection path in the physical topology model constructed in S2 is used as the flow constraint of the watershed algorithm: the traditional watershed algorithm allows water to diffuse to any adjacent pixel, while the improved algorithm strictly restricts water to flow along the physical connection path. In the specific implementation, a list of neighborhoods allowed to flow is established for each pixel, which is determined by the physical connection path determined in S2, rather than simply using the geometric 8-neighborhood or 4-neighborhood. For different types of LCD screens, the physical connection path has different characteristics: the physical connection in traditional TFT-LCD is mainly along the gate line and data line direction; OLED display screens may have more complex connection topology. In the algorithm implementation, the flow probability between adjacent pixels without physical connection is set to zero to ensure that water does not cross the physically isolated area.

[0062] The gradient value of the watershed algorithm is set according to the gradient weight matrix containing physical constraints generated in S2: the traditional watershed algorithm uses the image grayscale gradient as the terrain height, while the improved algorithm uses the gradient weight matrix containing physical constraints as the terrain height map. This gradient weight matrix has integrated the physical structure information of the LCD screen, so that the gradient value of the area that should be physically connected is lower, while the gradient value of the area that should be physically separated is higher. When used specifically, the gradient weight matrix is ​​converted into a terrain height map, and the area with low gradient values ​​corresponds to the terrain low points (reservoirs), and the area with high gradient values ​​corresponds to the terrain high points (ridges). In actual implementation, the gradient weight matrix is ​​usually normalized to ensure that the value range is within the range of [0, 1] or [0, 255] to facilitate subsequent processing.

[0063] Set the initial water level and use the pixel with the lowest gradient value as the initial immersion point: In each bad pixel location area to be processed, scan the gradient weight matrix containing physical constraints to find the set of pixels with the lowest gradient value. Use these lowest points as the initial immersion points, and usually assign a unique regional label to each initial immersion point as the seed point of the watershed algorithm. If multiple clearly separated gradient lowest points are detected in a single bad pixel area, they may correspond to multiple bad pixel sources, each of which serves as an independent initial immersion point. Initialize the priority queue data structure, sort all initial immersion points by their gradient values ​​and queue them to prepare for the subsequent immersion process.

[0064] During the immersion process, the water flow diffusion path is controlled according to the pixel connection weight matrix: for the path with connection weight N1 (physical distance less than the threshold and shared drive line), the diffusion priority is set to P1 (for example, the value is 9); for the path with connection weight N2 (physical distance less than the threshold but not shared drive line), the diffusion priority is set to P2 (for example, the value is 3); for the path with connection weight N3 (physical distance exceeds the threshold), the diffusion priority is set to P3 (for example, the value is 1); the relationship of P1>P2>P2 is satisfied to ensure that the algorithm preferentially diffuses along the physical connection path of the shared drive line. In terms of implementation, the diffusion rules of the traditional watershed algorithm are modified, and the candidate diffusion pixels are sorted according to the diffusion priority, and the diffusion pixels of the high priority path are processed first; the specific diffusion process adopts the priority queue data structure, and the priority of the elements in the queue is jointly determined by the pixel gradient value and the path diffusion priority. The formula for calculating the priority is: diffusion priority = pixel gradient value / path priority coefficient, so that the diffusion direction with a high path priority corresponds to a lower comprehensive priority value and will be processed first.

[0065] The driving line fault model constructed in S2 is used to predict the potential bad pixel diffusion direction, and the regional growth strategy of the watershed algorithm is adjusted: the abnormal pixel distribution pattern in the bad pixel location area is analyzed to identify the pixel anomalies distributed along a specific driving line. In the specific implementation, the number of pixels overlapping with each driving line in the bad pixel area is counted. When the abnormal pixel ratio on a driving line exceeds the preset threshold (usually 40%), it is determined to be an anomaly distributed along the driving line. When a pixel anomaly distributed along a specific driving line is detected, the gradient value weight between other pixel points on the driving line is reduced in the physically constrained gradient weight matrix, usually by 30%-50% of the original value. This will reduce the possibility of forming a watershed line between pixel points on the driving line. According to the physical routing structure of the driving line, the diffusion priority coefficient of the driving line direction is calculated. The specific method is to define a vector along the driving line direction and calculate the angle between the current diffusion direction and the driving line direction. The smaller the angle, the higher the priority coefficient. The calculated diffusion priority coefficient is applied to the immersion process of the watershed algorithm and multiplied by the original path priority value to further improve the diffusion priority along the driving line direction. This dual adjustment of gradient value weight and diffusion priority makes the watershed algorithm tend to classify abnormal pixels on the same driving line into the same region when forming regional boundaries, avoiding over-segmentation from a technical principle. The adjusted region growth strategy is applied to the entire immersion process of the watershed algorithm to continuously optimize the final segmentation result.

[0066] When the water level rises to a preset height, the minimum pixel distance between adjacent sub-regions is detected, and whether to establish a watershed line is determined based on the minimum pixel distance and the physical connection relationship: the minimum pixel distance between each sub-region formed during the immersion process is calculated, and the Euclidean distance between the two nearest pixels in the two regions is usually used as a metric. When the minimum pixel distance between adjacent sub-regions is less than a preset threshold (usually 3-5 pixels), their physical connection relationship is further analyzed to prevent over-segmentation. The position information of the pixels of the adjacent sub-regions in the two-dimensional pixel coordinate system is obtained to determine their exact position in the physical space. According to the position information, the drive line layout data is queried to obtain the connection relationship between the pixels in the adjacent sub-regions and the gate lines and data lines to determine whether they share the same drive line. At the same time, according to the position information and the drive line connection relationship, the physical connection path data is queried to determine whether there is a direct physical path between adjacent sub-regions. When adjacent sub-regions share the same drive line and there is a direct physical path on the physical connection path, no watershed line is established, but they are regarded as different parts of the same bad pixel area. This is a key mechanism to prevent over-segmentation. When adjacent sub-areas do not share the same driving line or do not have a direct physical access on the physical connection path, a watershed line is established to divide them into different bad pixel areas.

[0067] According to the established watershed line, the pixel groups are analyzed for continuity and merged to form the final segmentation result: for the identified pixel groups that share the same drive line, their spatial continuity is analyzed in the two-dimensional pixel coordinate system. The pixel groups that are continuously distributed and not separated by the watershed line are given the same region label, indicating that they belong to the same bad pixel area. The continuity analysis usually adopts the connected component labeling algorithm, but the constraints of the physical connection path need to be considered. Only physically connected pixels are considered continuous. The pixel groups with the same region label are merged to form multiple independent bad pixel sub-regions, each of which corresponds to a possible bad pixel or a group of related bad pixels. For each independent bad pixel sub-region formed, its boundary coordinates, area, shape characteristics and the driving line information involved are recorded to provide a basis for the subsequent bad pixel classification and inducement analysis. The final output segmentation result is presented in the form of a region labeling map, where different region labeling values ​​represent different bad pixel sub-regions, and the region with a value of 0 represents a non-bad pixel region.

[0068] S4, using the pre-built bad pixel feature knowledge base to perform feature analysis on the segmentation results, determine the bad pixel name and bad pixel cause, and obtain the detection results, including: the system receives the segmentation result image with region markings output by the S3 watershed algorithm, where each independent bad pixel sub-region has a unique marking value. The received segmentation results are preprocessed, including noise filtering and boundary smoothing. Usually, tiny areas with an area smaller than a threshold (such as 3-5 pixels) are filtered out to eliminate possible false detections. A data structure is established for each independent bad pixel sub-region, recording its marking value, the list of pixel coordinates contained, the list of boundary pixel coordinates and other basic information in preparation for subsequent analysis. An index table for bad pixel sub-regions is established to facilitate rapid positioning and retrieval of information on specific areas during subsequent analysis.

[0069] Calculate the geometric characteristic parameters of each area based on the independent bad pixel sub-area: Area: Calculate the total number of pixels contained in each bad pixel sub-area. For high-resolution displays, the number of pixels is usually converted into physical area (square millimeters). Perimeter: Calculate the number of boundary pixels of the bad pixel sub-area. Boundary pixels are defined as pixels where at least one adjacent pixel does not belong to the area. For more accurate perimeter calculation, an 8-connectivity-based boundary tracing algorithm is usually used. Aspect ratio: Determine the minimum enclosing rectangle of the bad pixel sub-area and calculate the ratio of its major axis to its minor axis. The aspect ratio is an important parameter for distinguishing between point-like and linear bad pixels. Usually, the aspect ratio of linear bad pixels is >3.

[0070] Shape complexity: Calculates shape complexity indicators. Common methods include: Circularity: 4π×area / perimeter², a perfect circle is 1, and the more irregular the shape, the smaller the value; Rectangularity: area of ​​the region / area of ​​the smallest enclosing rectangle, a perfect rectangle is 1; Fractal dimension: calculated through the relationship between the length of the boundary and the change in the measurement scale.

[0071] Distribution density: Analyze the distribution density of pixels in the area, including: pixel density: number of bad pixels in the area / area of ​​the smallest circumscribed rectangle; density gradient: rate of change of pixel density from the center to the edge of the area; void rate: number of non-bad pixels in the area / total area of ​​the area.

[0072] According to the two-dimensional pixel coordinate system, analyze the position characteristics of each bad pixel sub-area on the LCD screen: calculate the centroid coordinates of the bad pixel sub-area, that is, the average value of the pixel coordinates in all areas, as the representative point of the area position. Calculate the distance of the bad pixel sub-area relative to the four sides of the screen to determine whether it is an edge bad pixel (usually the distance is less than 5% of the screen size). Edge bad pixels are usually related to frame pressure, poor welding and other problems. Analyze the distance and direction of the bad pixel relative to the driver IC position, which is particularly important for determining bad pixels caused by driver IC failure.

[0073] Calculate the relative position relationship between the current bad pixel sub-area and other bad pixel areas, including: the nearest bad pixel distance: the minimum distance between the current area and the nearest other bad pixel area; spatial distribution pattern: whether multiple bad pixel areas present a specific geometric arrangement (such as linear arrangement, grid arrangement); directional consistency: whether multiple bad pixels are arranged in the same direction; record the distribution of bad pixels in different functional areas of the display screen (such as menu area, signal line area, display main area), which is of great significance for the judgment of specific types of bad pixels.

[0074] According to the drive line layout data, analyze the correlation between the bad pixel sub-area and the drive line: check whether the bad pixel sub-area overlaps with a specific gate line or data line, and calculate the degree of overlap (number of overlapping pixels / total number of pixels in the area). When the degree of overlap exceeds the preset threshold (usually 60%), it is determined that the bad pixel is highly correlated with the drive line. Analyze the ductility of the bad pixel area in the direction of the drive line, and calculate the proportion of the projection length of the bad pixel area in the direction of the drive line to the total length of the line. Detect whether the bad pixel area spans multiple parallel drive lines. If so, it may be related to the cross-point fault of the drive line. Analyze the distribution pattern of multiple separated bad pixel areas on the same drive line to determine whether there is a regular fault caused by the attenuation of the drive signal transmission. According to the transmission direction of the drive signal, analyze the distribution characteristics of the bad pixel area in the signal flow direction, which is particularly important for judging signal transmission faults. Record the drive line identification, type and physical characteristics related to the bad pixel to provide a basis for subsequent cause analysis.

[0075] Bad pixels are classified based on geometric feature parameters, position features and drive line correlation: The criteria for determining point-like bad pixels: the bad pixel area is composed of N adjacent pixels (usually N<10); the aspect ratio is close to 1 (usually <1.5); the shape complexity index is close to a circle or square; the pixels in the area are evenly distributed without obvious directionality; typical point-like bad pixels include single bright / dark spots, stuck spots (fixed color spots), etc.

[0076] The criteria for determining linear bad pixels: the bad pixel area obviously extends along the direction of the drive line; the aspect ratio is large (usually >3); the overlap with the specific drive line is high (>60%); the ductility of the area in one direction is much greater than that in other directions; typical linear bad pixels include gate line open circuit, data line short circuit, abnormal drive line signal transmission, etc.

[0077] The criteria for determining blocky bad pixels: the bad pixel area is a collection of pixels distributed in a rectangular or irregular area; the area is large (usually >100 pixels); and multiple drive lines may be affected at the same time.

[0078] Further subdivided according to the shape: rectangular block: high rectangularity (>0.8), regular boundaries; irregular block: complex shape, irregular boundaries; typical block bad pixels include TFT array defects, backlight leakage, poor liquid crystal injection, etc.; for composite bad pixels (bad pixels with multiple characteristics at the same time), multi-feature comprehensive judgment is used to determine their main type and secondary type.

[0079] Match the classification results with the characteristic parameters in the bad pixel feature knowledge base: The bad pixel feature knowledge base is a pre-built database containing characteristic parameters, cause analysis and identification rules of various types of bad pixels. This knowledge base is usually built based on a large number of historical bad pixel cases and expert experience.

[0080] For each bad pixel sub-region feature parameter combination, multi-dimensional feature matching is performed in the knowledge base. Common matching algorithms include: distance-based nearest neighbor matching: calculate the Euclidean distance between the current bad pixel feature vector and the feature vectors of each type of bad pixel in the knowledge base, and select the type with the smallest distance; similarity-based fuzzy matching: use indicators such as cosine similarity to calculate the similarity of feature vectors; rule-based expert system matching: use predefined IF-THEN rules for judgment; the matching process usually adopts a weighted method to assign different weights to different feature parameters to reflect their importance in the bad pixel type judgment. Evaluate the confidence of the matching results. When the confidence is lower than the threshold, manual confirmation or further analysis may be required. Based on the matching results, determine the specific type and name of the bad pixel, such as "single pixel dark spot", "gate line open circuit", "liquid crystal leakage block", etc.

[0081] Based on the identified bad pixel type and distribution characteristics, combined with the cause analysis rules in the knowledge base, the possible causes of bad pixel formation are inferred: for each bad pixel type, the knowledge base predefines a list of possible causes and their associated feature patterns. By comparing the feature pattern of the current bad pixel with the feature pattern associated with the cause, the probability score of each possible cause is calculated. The cause analysis process considers a variety of factors, including: the type and location characteristics of the bad pixel; the correlation between the bad pixel and the drive circuit; the spatial relationship between multiple bad pixels; the change pattern of pixel values ​​in the bad pixel area; and the main causes of similar bad pixels in historical statistical data.

[0082] Common causes of bad pixels include: manufacturing process defects: such as TFT process defects, mask alignment errors, poor lithography, etc.; material problems: such as impurities in liquid crystal materials, polarizer defects, poor color filters, etc.; electrical failures: such as damage to the driver IC, open or short circuits in the gate line / data line, etc.; mechanical damage: such as external force extrusion, panel bending, thermal stress, etc.; aging degradation: such as backlight attenuation, directional failure of liquid crystal molecules, etc. Generate a cause analysis report, list possible causes and their probability ranking, and provide a reference for production process improvements.

Claims

1. A method for detecting bad pixels on a liquid crystal display, characterized in that: include: S1, obtaining the bad pixel location area of ​​the LCD screen according to a preset test plan; S2, constructing a physical topology model of the LCD screen according to the physical parameters of the LCD screen; S3, using the improved watershed algorithm to segment the bad pixel location area to obtain the segmentation result; S4, using a pre-built bad pixel feature knowledge base to perform feature analysis on the segmentation result, determine the bad pixel name and bad pixel cause, and obtain the detection result; Among them, the improved watershed algorithm is used to segment the bad pixel location area, including: By using the physical topology model, the flow rules based on the physical connection path are adopted so that the watershed algorithm can only diffuse along the physical connection path to obtain the segmentation result.

2. The method for detecting bad pixels of a liquid crystal display according to claim 1, characterized in that: The physical parameters include RGB pixel arrangement, sub-pixel structure and drive circuit layout; The bad pixel feature knowledge base includes feature parameters, cause analysis and identification rules for various types of bad pixels.

3. The method for detecting bad pixels of a liquid crystal display according to claim 2, characterized in that: S1, obtaining the bad pixel location area of ​​the LCD screen according to the preset test plan, including: Determine the standard gradient value corresponding to each pixel point in the liquid crystal screen according to the preset test scheme, and generate a standard gradient value matrix; the standard gradient value represents the grayscale gradient value corresponding to the pixel point under the preset normal working conditions; Acquire a test image of the liquid crystal screen under test conditions; the test image represents a display image of the liquid crystal screen under the test mode; Construct the actual gradient value matrix based on the test image; By comparing the standard gradient value matrix with the actual gradient value matrix, the bad pixel location area of ​​the LCD screen is determined.

4. The method for detecting bad pixels of a liquid crystal display screen according to claim 3, characterized in that: S2, constructing a physical topology model of the LCD screen based on the physical parameters of the LCD screen, including: Obtain the RGB pixel arrangement of the LCD screen, determine the physical connection relationship between adjacent pixels, and the position relationship between adjacent pixels; A two-dimensional pixel coordinate system is established based on the physical connection relationship between adjacent pixels and the position relationship between adjacent pixels; Determine the physical connection path at the sub-pixel level according to the physical boundaries of the sub-pixels of the LCD screen; Obtain the layout data of the driving circuit of the LCD screen, determine the connection relationship between the pixel points and the gate lines and data lines according to the physical wiring structure of the driving circuit, and identify the pixel point groups sharing the same driving circuit by tracing the transmission path of the driving signal; Construct a pixel connection weight matrix based on the RGB pixel arrangement, sub-pixel and drive line layout data; Generate a gradient weight matrix containing physical constraints according to the pixel connection weight matrix and the standard gradient value matrix; According to the drive line layout data, the drive line fault model is established through the transmission characteristics and attenuation law of the drive signal on the line; A physical topology model of the LCD screen is constructed based on the two-dimensional pixel coordinate system, the physical connection path at the sub-pixel level, the pixel connection weight matrix and the drive line fault model.

5. The method for detecting bad pixels of a liquid crystal display according to claim 4, characterized in that: Construct a pixel connection weight matrix, including: Get the physical distance between pixels and the sharing of driving lines; For the pixels whose physical distance is less than the threshold and share the driving line, the connection weight is set to N1; For pixels whose physical distance is less than the threshold but do not share a driving line, the connection weight is set to N2; The connection weight is set to N3 for pixels whose physical distance exceeds the threshold.

6. The method for detecting bad pixels of a liquid crystal display screen according to claim 5, characterized in that: N1 is greater than N2, and N2 is greater than N3.

7. The method for detecting bad pixels of a liquid crystal display according to claim 4, characterized in that: S3, using the improved watershed algorithm to segment the bad pixel location area, and obtain the segmentation results, including: The bad pixel location area determined in S1 is used as the input area of ​​the watershed algorithm; The physical connection paths in the physical topology model constructed in S2 are used as flow constraints for the watershed algorithm; Set the gradient value of the watershed algorithm according to the gradient weight matrix containing physical constraints generated in S2; Set the initial water level and take the pixel with the lowest gradient value as the initial immersion point; During the immersion process, the water flow diffusion path is controlled according to the pixel connection weight matrix. The diffusion priority of the path with connection weight N1 is set to P1, the diffusion priority of the path with connection weight N2 is set to P2, and the diffusion priority of the path with connection weight N3 is set to P3. The driving line fault model in S2 is used to predict the potential diffusion direction of bad points and adjust the regional growth strategy of the watershed algorithm; When the water level rises to a preset height, the minimum pixel distance between adjacent sub-areas is detected, and a watershed line is established based on the minimum pixel distance; According to the established watershed line, for the identified pixel point groups sharing the same driving line, the corresponding continuity is analyzed in the two-dimensional pixel coordinate system, and the pixel groups that are continuously distributed and not separated by the watershed line are given the same area label; Pixel groups with the same region label are merged to form multiple independent bad pixel sub-regions as the segmentation result.

8. The method for detecting bad pixels of a liquid crystal display screen according to claim 7, characterized in that: P1 is greater than P2, and P2 is greater than P3.

9. The method for detecting bad pixels of a liquid crystal display according to claim 7, characterized in that: When the water level rises to a preset height, the minimum pixel distance between adjacent sub-areas is detected, and a watershed line is established based on the minimum pixel distance, including: Calculate the minimum pixel distance between each sub-area formed during the immersion process; When the minimum pixel distance between adjacent sub-regions is less than a threshold, obtaining position information of pixel points of the adjacent sub-regions in the two-dimensional pixel coordinate system constructed in S2; According to the position information, the connection relationship between the pixel points in the adjacent sub-area and the gate lines and the data lines in the driving circuit layout data in S2 is obtained; According to the position information and the connection relationship between the pixel point and the gate line and the data line, the physical connection path of the pixel point in the adjacent sub-area determined in S2 is obtained, and the physical connection path is used to determine whether there is a direct physical path between the adjacent sub-areas; When adjacent sub-regions share the same driving line and there is a direct physical access on the physical connection path, no watershed line is established; A watershed line is established when adjacent sub-regions do not share the same drive line or do not have direct physical access on a physically connected path.

10. A liquid crystal display screen bad pixel detection system, characterized in that: include: At least one processing unit; used to execute instructions to implement the method for detecting bad pixels on a liquid crystal display screen as described in any one of claims 1 to 9.

Images