Medical image display system based on color calibration

Abstract

The invention relates to the technical field of image processing, in particular to a medical image display system based on color calibration, and the system comprises a calibration tensor generation module which generates a three-dimensional calibration input tensor; the closed-loop feedback calibration module is used for inputting the three-dimensional calibration input tensor into a pre-trained multi-modal adversarial compensation network and generating a pixel-level aging-environment coupling compensation coefficient matrix and a color gamut anchor point coordinate; and the compensation execution module is used for writing the compensated quantized color instruction set into a redundant driving unit of the display panel to realize cooperative offset of equipment aging and environmental interference. According to the method, the color rendition error in the high-resolution medical image can be effectively reduced, the color stability is improved, and the method is suitable for real reproduction of high-precision diagnosis scenes such as endoscope and CT pseudo-color images.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of image processing, and in particular to a medical image display system based on color calibration. BACKGROUND

[0002] As an important information carrier for clinical diagnosis and intraoperative navigation, medical images have very high requirements for the color restoration accuracy and stability of their display systems. Currently, medical display devices generally use high-resolution panels such as liquid crystal (LCD), organic light-emitting diode (OLED), or quantum dot light-emitting diode (QLED). However, in the long-term operation process, there is an irreversible brightness decay problem for each sub-pixel of the panel, and the aging rates of different materials are inconsistent in the red, green, and blue channels, leading to increasingly serious color imbalance.

[0003] On the other hand, the environmental light sources in medical scenarios such as operating rooms and radiology reading rooms are complex and variable, and their reflected and scattered light can significantly interfere with the display content. Especially when displaying low-contrast images (such as angiography and pathological sections), problems such as false color, color deviation, and reduced contrast often occur, which seriously affect the accuracy of diagnosis and visual comfort. Although some existing display systems integrate brightness sensors or ambient light sensors, most of them only perform rough adjustment on the overall brightness and cannot achieve fine compensation for specific display areas and specific pixels.

[0004] In addition, traditional color calibration techniques are mostly based on preset color lookup tables (LUTs) or offline correction algorithms, which are difficult to dynamically adapt to the coordinated changes of device aging and environmental interference. Moreover, the compensation parameters are usually generated by single-channel models, which lack deep perception of image content characteristics, leading to defects such as color gamut drift, inter-frame inconsistency, and response lag in the compensated images, especially in dynamic medical image sequences. SUMMARY

[0005] The present application provides a medical image display system based on color calibration, which realizes sub-pixel level dynamic color correction through a multi-modal color compensation mechanism of device aging state, environmental light interference characteristics, and image content color gamut distribution, combined with intelligent feedback and hardware driving collaboration, to ensure the color accuracy and system stability of medical images in complex application scenarios.

[0006] A medical image display system based on color calibration, comprising:

[0007] A calibration tensor generation module: synchronously acquires real-time aging fingerprint spectrum, environmental light interference matrix, and color gamut feature topology of the image to be displayed, and generates a three-dimensional calibration input tensor, wherein the aging fingerprint spectrum includes the brightness decay gradient distribution of each sub-pixel of the display panel.

[0008] The closed-loop feedback calibration module inputs the three-dimensional calibration input tensor into a pre-trained multi-modal adversarial compensation network to generate a pixel-level aging-environment coupling compensation coefficient matrix and color gamut anchor point coordinates, and the compensation coefficient matrix and the color gamut anchor point coordinates form a closed-loop feedback calibration path.

[0009] The compensation execution module performs sub-pixel level color remapping on the input image based on the closed-loop feedback calibration path, writes the compensated quantized color instruction set into the redundant driving unit of the display panel, and realizes the synergistic offset of device aging and environmental interference.

[0010] Optionally, the calibration tensor generation module specifically comprises:

[0011] The aging fingerprint map construction unit detects the light-emitting efficiency decay value of the red, green and blue sub-pixels pixel by pixel in a frame scanning manner through the micro photoelectric sensor array embedded in the display panel, and constructs the aging fingerprint map based on the spatial distribution of the decay value.

[0012] The ambient light interference matrix generation unit synchronously collects the wavelength-intensity interference spectrum of ambient light by using a multi-spectral ambient light probe, extracts the coherence features of light intensity of each waveband by Fourier transform, and generates an ambient light interference matrix including spectral shift and intensity modulation factor.

[0013] The color gamut feature topology analysis unit analyzes the color gamut features of the to-be-displayed image, extracts the topological correlation of the chrominance coordinates and brightness values of each color block in the CIE xyY space, and constructs a color gamut feature topology graph.

[0014] The three-dimensional calibration tensor construction unit tensor splices the spatial decay distribution of the aging fingerprint map, the frequency domain feature vector of the ambient light interference matrix, and the chrominance correlation weight of the color gamut feature topology graph to obtain a three-dimensional calibration input tensor.

[0015] Optionally, in the aging fingerprint map, the brightness decay gradient distribution of each sub-pixel is expressed as:

[0016] ;

[0017] wherein, represents the normalized brightness decay gradient of the sub-pixel type at the th position, represents the brightness value of the sub-pixel type at the initial factory, represents the brightness value of the th sub-pixel at the current moment, represents a nonlinear aging coefficient related to the cumulative working time , and the aging fingerprint map is formally expressed as: .

[0018] Optionally, the three-dimensional calibration input tensor is formed along a channel dimension, and has a size of HxWx(N+M+P), where H and W are display panel resolutions, N is a number of aging parameter channels, M is an ambient light feature dimension, and P is a color gamut topology correlation dimension.

[0019] Optionally, the multi-modal adversarial compensation network includes a generator branch and a compensation verifier branch, and the closed-loop feedback calibration module specifically includes:

[0020] a compensation coefficient generation unit: the generator branch extracts aging-environment coupling features in the three-dimensional calibration input tensor using a depth separable convolution, and generates a pixel-level aging-environment coupling compensation coefficient matrix including red, green, and blue sub-pixel compensation gain values;

[0021] a color gamut anchor point extraction unit: the compensation verifier branch synchronously processes color gamut feature topologies in the three-dimensional calibration input tensor, extracts dynamic distribution features of an image color gamut through an adaptive spatial pooling layer, and outputs a color gamut anchor point coordinate set in a CIE LAB color space;

[0022] a generator-verifier adversarial unit: the generator branch and the compensation verifier branch form a dynamic game mechanism through adversarial training: the compensation coefficient matrix is applied to a test image in a simulated aging environment to generate a compensated image, the compensated image is input into the compensation verifier branch for color gamut anchor point matching degree verification, and if a color gamut anchor point coordinate offset exceeds a preset threshold, the convolution kernel weight of the generator branch is corrected in reverse;

[0023] a closed-loop calibration path construction unit: a color gamut anchor point coordinate generated in a current frame is analyzed in space-time correlation with a color gamut feature topology of a next frame image, a color gamut anchor point drift vector is generated and fed back to the generator branch, and the spatial distribution weight of the compensation coefficient matrix is adjusted in real time;

[0024] a compensation coefficient nonlinear mapping unit: the pixel-level aging-environment coupling compensation coefficient matrix and the color gamut anchor point coordinate form a nonlinear mapping relationship.

[0025] Optionally, the adaptive spatial pooling layer extracts dynamic distribution features of an image color gamut, and outputs a color gamut anchor point coordinate set in a CIE LAB color space specifically includes:

[0026] After receiving the color gamut feature topology map in the calibration input tensor, the compensation verifier branch enhances the key color response areas in the color gamut channel through the channel attention mechanism to obtain the enhanced color gamut feature map. The enhanced color gamut feature map is input into the adaptive spatial pooling layer. The adaptive spatial pooling layer automatically adjusts the pooling window size and sliding step size based on the color distribution density of different regions in the current image, so that the pooling operation adaptively performs fine-grained aggregation on high color variation areas and performs large-scale compression on low-variance areas, preserving the structural information of color transition areas and edges.

[0027] The compact color gamut representation obtained after pooling is input to the feature compressor, converted to the CIE LAB color space, and clustering and color center extraction are performed on each local area. The representative color center is extracted as the color gamut anchor point, which reflects the main color distribution position of the current frame image in the perceptual space.

[0028] Optionally, the spatiotemporal correlation analysis in the closed-loop calibration path construction unit specifically includes:

[0029] When performing color gamut anchor point drift analysis, the color gamut anchor point coordinate set extracted from the current frame image is spatially aligned with the color gamut feature topology map in the next frame image to ensure that the corresponding relationship between the color gamut anchor points in the display area between the two frames is consistent. The color distribution characteristics between the anchor points are then used for feature matching to determine the optimal corresponding target for the current frame anchor point in the next frame.

[0030] For each pair of successfully matched color gamut anchor points, the direction and magnitude of color drift are determined based on their coordinate differences in the CIE LAB space and their position changes in the image space. A cross-frame anchor point drift vector is generated and fed back to the compensation network to adjust the compensation parameter distribution of subsequent frames, thereby maintaining color consistency between consecutive frames.

[0031] Optionally, in the compensation coefficient matrix nonlinear mapping unit, a full sigmoid nonlinear mapping relationship between the pixel-level aging-environment coupling compensation coefficient matrix and the aging fingerprint attenuation gradient and the offset distance between the current color gamut anchor point and the standard color gamut is expressed as:

[0032] ;in, Indicates the The aging fingerprint attenuation gradient of the sub-pixel at the position, Indicates the corresponding color channel The Euclidean distance between the color gamut anchor point and the center of the standard color gamut, Represents the mapping sensitivity factor that is dynamically optimized during adversarial training.

[0033] Optionally, the compensation execution module specifically includes:

[0034] Brightness correction calculation unit: the pixel-level aging-environment coupling compensation coefficient matrix is convolved with the original pixel value of the current input image to generate a compensated RGB three-channel brightness correction value;

[0035] Boundary constraint and coordinate remapping unit: according to the color gamut anchor point coordinate set, a dynamic color gamut enclosure is constructed in the CIE LAB color space, the modified pixel value is subjected to chroma boundary constraint, and sub-pixel level color coordinate remapping is realized by Lagrange interpolation algorithm;

[0036] Color instruction writing unit: the remapped color data is converted into a color instruction set, the color instruction set includes the driving pulse width modulation parameter and voltage gradient compensation value of each sub-pixel, and is written in real time through the standby TFT circuit in the redundant driving unit built in the display panel.

[0037] Optionally, the compensation execution module is also provided with a double buffer register, specifically including a double buffer register group arranged in the redundant driving unit;

[0038] When the first register group executes the current frame compensation instruction, the second register group synchronously receives the closed-loop feedback calibration parameters of the next frame, realizing low-delay switching of aging compensation and environmental interference cancellation.

[0039] The beneficial effects of the present application are:

[0040] The present application comprehensively perceives the key factors affecting color accuracy by constructing a three-dimensional calibration input tensor containing display device aging fingerprint map, environmental light interference matrix and image color gamut feature topology, inputs into a multi-modal adversarial compensation network, realizes collaborative modeling and pixel-level compensation of display aging and environmental changes, compared with the traditional method of only based on brightness or color temperature parameter for single channel correction, the color restoration error in high resolution medical image of the present application can be effectively reduced, the color stability is improved, and the present application is suitable for high precision diagnosis scene such as endoscope, CT pseudo-color image real reproduction.

[0041] The generator branch of the present application can dynamically output aging-environment coupling compensation coefficients, and the verifier branch can extract color gamut anchor points through adaptive spatial pooling and verify the compensation effect in real time, and the two constitute a closed-loop optimization path through adversarial training mechanism and anchor point drift vector feedback, which not only can realize self-supervised optimization of each frame color remapping, but also can maintain color gamut space continuity between multiple images, overcome the color flicker and offset problem caused by device state fluctuation or frame difference in the prior art, and improve the matching degree.

[0042] The application introduces a redundant drive unit and a double-buffer register group in a compensation execution module, combines color instruction quantization and reverse compensation current injection mechanism, breaks through the limitation of traditional external algorithm correction, realizes hardware-level response to TFT aging, sub-pixel drive attenuation and line impedance drift, and especially in high-level medical display equipment, the structure can realize pulse width + voltage dual-channel regulation and control, and quickly switch to the standby channel when the main drive fails, ensuring image continuity and color accuracy. BRIEF DESCRIPTION OF DRAWINGS

[0043] In order to more clearly illustrate the technical solutions in the present application or prior art, the following will briefly introduce the drawings needed to be used in the embodiments or prior art description. Obviously, the drawings in the following description only illustrate the present application, and other drawings can also be obtained by those skilled in the art without any creative effort.

[0044] Figure 1 The figure is a schematic diagram of the system function module of the embodiment of the present application.

[0045] Figure 2 The figure is a schematic diagram of the calibration tensor generation module of the embodiment of the present application.

[0046] Figure 3 The figure is a schematic diagram of the closed-loop feedback calibration module of the embodiment of the present application.

[0047] Figure 4 The figure is a schematic diagram of the compensation execution module of the embodiment of the present application. DETAILED DESCRIPTION

[0048] The present application will be described in detail below with reference to the drawings and specific embodiments. For some known technologies, other alternative ways can also be adopted by those skilled in the art to implement them; and the drawings are only used to more specifically describe the embodiments, and are not intended to specifically limit the present application.

[0049] As shown in the figure, a medical image display system based on color calibration includes: Figures 1-4

[0050] The calibration tensor generation module synchronously acquires the real-time aging fingerprint spectrum of the display device, the ambient light interference matrix and the color gamut feature topology of the image to be displayed, generates a three-dimensional calibration input tensor, and the aging fingerprint spectrum includes the brightness attenuation gradient distribution of each sub-pixel of the display panel.

[0051] The closed-loop feedback calibration module inputs the three-dimensional calibration input tensor into the pre-trained multi-modal adversarial compensation network, generates a pixel-level aging-environment coupling compensation coefficient matrix and a color gamut anchor point coordinate, and the compensation coefficient matrix and the color gamut anchor point coordinate form a closed-loop feedback calibration path. ​

[0052] Compensation execution module: Based on a closed-loop feedback calibration path, it performs sub-pixel color remapping on the input image and writes the compensated quantized color instruction set into the redundant drive units of the display panel to achieve synergistic offset of device aging and environmental interference.

[0053] The calibration tensor generation module specifically includes:

[0054] Aging fingerprint map construction unit: Through the micro-photoelectric sensor array embedded in the display panel, the luminous efficiency attenuation value of the red, green, and blue sub-pixels is detected pixel by pixel in a frame scanning manner, and the aging fingerprint map is constructed based on the spatial distribution of the attenuation value. The brightness attenuation gradient distribution of each sub-pixel is calculated by the following formula:

[0055] ;in, Indicates the Sub-pixel type at position The normalized brightness attenuation gradient of Indicates sub-pixel type The initial brightness value at the factory, Indicates the current moment The brightness value of the sub-pixel, Indicates and accumulates working time The related nonlinear aging coefficient is expressed as follows: ,in, is the aging constant of the display panel material. Assume that the display panel resolution is , then the aging fingerprint can be formally expressed as: .

[0056] Ambient light interference matrix generation unit: uses a multi-spectral ambient light probe to synchronously collect the wavelength-intensity interference pattern of ambient light, extracts the coherence characteristics of the light intensity in each band through Fourier transform, and generates an ambient light interference matrix including spectral offset and intensity modulation factor;

[0057] Assume that the wavelength-intensity sequence of the ambient light collected by the multi-spectral ambient light probe is:

[0058] ; For the The central wavelength of the band, is the true spectral intensity, To measure the noise interference term, is the total number of sampling bands (for example, the visible light region of 400-700nm is divided into 32~64 segments), and the sampling sequence on the wavelength axis is Perform a one-dimensional fast Fourier transform (FFT):

[0059] ; wherein is the frequency index (corresponding to the band periodic component), is the complex amplitude of the th frequency component; define the coherence factor and spectral shift as: ; wherein is the th normalized intensity factor of the frequency domain component, is the maximum response wavelength (i.e. shift center) corresponding to the principal component, represents the real part of the complex spectrum; the final ambient light interference matrix can be represented as:

[0060] ; wherein is the number of principal frequency components selected, and each row of the matrix corresponds to a band interference feature (intensity factor + shift center).

[0061] Color gamut feature topology analysis unit: perform color gamut feature analysis on the image to be displayed, extract the topological correlation relationship of the chromaticity coordinates and luminance values of each color block in the image in the CIE xyY space, and construct a color gamut feature topology graph, which specifically includes converting pixel point RGB to CIExyY coordinates:

[0062] ;

[0063] For any pixel in the image to be displayed, the RGB value is converted to XYZ tristimulus value, wherein, : RGB to XYZ color space conversion matrix (according to sRGB or D65 standard), represents the CIE1931 tristimulus value, is the normalized RGB input value. Then calculate the xyY coordinates:

[0064] ;

[0065] Cluster the triplet of pixel points of the entire image to obtain color block (cluster) anchor points; construct a topology graph between color blocks , wherein, , each vertex represents a color block, , the edge weight is defined as the color difference between color blocks + spatial adjacent weight:

[0066] ; wherein, represents the chrominance-luminance vector of the center of the th color block, is the image space distance between the centers of the color block anchor points, For the Gaussian adjustment factor of color difference and spatial distance, the finally constructed color gamut feature topology is expanded in the form of a feature map in the channel splicing input tensor. .

[0067] Three-dimensional calibration tensor construction unit: fuse the following three types of features into tensors, and splice along the channel dimension to construct a three-dimensional calibration input tensor with a size of .

[0068] Spatial decay distribution features of aging fingerprint maps (channel dimension number );

[0069] Frequency domain feature vectors of ambient light interference matrix (channel dimension number );

[0070] Chromaticity correlation weights of color gamut feature topology (channel dimension number );

[0071] wherein, , are the vertical and horizontal resolutions of the display panel, is the number of aging parameter dimensions, and the number of decay maps for each sub-pixel type, is the ambient light frequency domain feature dimension, which is determined according to the probe band resolution, is the correlation dimension of the color gamut topology, which is determined according to the image blocking and chroma clustering results.

[0072] The closed-loop feedback calibration module specifically comprises:

[0073] Compensation coefficient generation unit of generator branch: the multi-modal adversarial compensation network includes a generator branch and a compensation verifier branch, wherein the generator branch receives a three-dimensional calibration input tensor, extracts aging channel-environment channel coupling features in the tensor using a depth separable convolution structure, and generates a pixel-level compensation coefficient matrix with a size of : ; wherein, represents the compensation gain value of the sub-pixel type at position , is the resolution dimension of the display panel.

[0074] Color gamut anchor point extraction unit of verifier branch: the compensation verifier branch synchronously processes the color gamut feature topology in the calibration input tensor, extracts color gamut dynamic distribution features of different image regions through an adaptive spatial pooling layer, and outputs a color gamut anchor point coordinate set in the CIELAB color space:

[0075] ; wherein, is the number of color gamut anchor points extracted in the image, CIELAB coordinates representing the first color gamut anchor point.

[0076] The adaptive spatial pooling layer is as follows:

[0077] After the compensation verifier branch receives the color gamut feature topology in the calibration input tensor, it first enhances the key color response area in the color gamut channel through the channel attention mechanism, and then inputs the enhanced color gamut feature map into the adaptive spatial pooling layer. Based on the color distribution density of different regions in the current image, the pooling window size and sliding step are automatically adjusted, so that the pooling operation can adaptively aggregate the high color change region in a fine-grained manner, and perform large-scale compression on the low change region, thereby retaining the structural information of the color transition region and edge. The compact color gamut representation obtained after pooling is sent to the feature compressor, converted to the CIE LAB color space, and clustered and analyzed for each local region to extract the color center. Finally, the representative color center is extracted as the color gamut anchor point, and the anchor point set can reflect the main color distribution position of the current frame image in the perception space, serving as a reference basis for subsequent compensation accuracy verification.

[0078] Generator-verifier adversarial unit: closed-loop training through dynamic game mechanism, specifically:

[0079] 1. Apply the compensation coefficient matrix to the test image in the simulated aging environment to generate the compensated image ;

[0080] 2. Input into the verifier branch to extract its color gamut anchor point set ;

[0081] 3. If any color gamut anchor point offset:

[0082] , trigger error backpropagation to update the convolution kernel weights of the generator branch, where is the color gamut anchor point offset tolerance threshold (1.5 CIE units), is the Euclidean distance.

[0083] Closed-loop calibration path construction unit: perform spatio-temporal correlation analysis on the anchor point coordinate set generated in the current frame image and the color gamut topology of the next frame image to calculate the anchor point drift vector :

[0084] ; feed the anchor point drift vector back to the generator branch to adjust the spatial distribution weight of the compensation coefficient matrix in real time in the next round of training.

[0085] The spatio-temporal correlation analysis is specifically: when performing anchor point drift analysis, first align the color domain anchor point coordinate set extracted from the current frame image with the color domain feature topology graph in the next frame image in spatial coordinates, ensure the correspondence relationship of the anchor points between the two frames on the display area is consistent, then use the color distribution features (such as chrominance value, brightness level and local adjacency) between the anchor points to perform feature matching, determine the optimal corresponding target of the current frame anchor point in the next frame. For each pair of successfully matched anchor points, according to the coordinate difference in the CIE LAB space, combined with the position change in the image space, the direction and amplitude of the color drift are comprehensively judged, so as to generate the anchor point drift vector across frames, which reflects the display offset trend caused by device aging or environmental change, and also embodies the influence of image content dynamic change on color domain distribution. Finally, the drift vector will be fed back to the compensation network for dynamically adjusting the compensation parameter distribution of the subsequent frame, realizing the maintenance of color consistency between continuous frames.

[0086] Pixel-level compensation coefficient nonlinear mapping unit: pixel-level aging-environment coupling compensation coefficient matrix and the decay gradient of the aging fingerprint atlas and the offset distance of the current color domain anchor point and the standard color domain satisfy the following nonlinear mapping relationship in sigmoid form:

[0087] ; wherein, represents the decay gradient of the sub-pixel at the position, represents the Euclidean distance between the color domain anchor point of the corresponding color channel and the center of the standard color domain, represents the mapping sensitivity factor dynamically optimized in the adversarial training.

[0088] The compensation execution module specifically includes:

[0089] Luminance correction calculation unit: perform sub-pixel convolution processing on the pixel-level aging-environment coupling compensation coefficient matrix and the original pixel value of the current input image to obtain the luminance correction value of each sub-pixel channel. Taking the red channel as an example, the correction calculation is:

[0090] ;

[0091] wherein, is the corrected red sub-pixel luminance value, represents the original red channel luminance, represents the aging-environment coupling compensation gain coefficient, a decay gradient of red sub-pixels in the aging fingerprint map, the Euclidean distance between the color gamut anchor point of the current red channel and the standard color gamut center, a very small constant for preventing division by zero, and other channels The same processing as above.

[0092] Boundary constraint and coordinate remapping unit: according to the extracted color gamut anchor point coordinate set, a minimum enclosure is constructed in the CIE LAB color space to limit the color gamut boundary range of the calibrated pixel value. For the corrected pixel value falling within the boundary range, Lagrange interpolation algorithm is used to interpolate and remap its chroma coordinates between anchor points to achieve sub-pixel level color accurate positioning and avoid excessive deviation or color jump.

[0093] Color instruction writing unit: quantize the pixel data processed by remapping into a color driving instruction set, which includes:

[0094] Pulse width modulation (PWM) parameters of each sub-pixel;

[0095] Compensation values corresponding to voltage gradients;

[0096] The above instructions are written into the standby TFT circuit of the redundant driving unit in the display panel in real time, and are updated frame by frame by the hardware controller.

[0097] Double buffer register (realize low delay switching):

[0098] Double buffer register groups are arranged in the redundant driving unit:

[0099] The first register group is used for real-time driving of the current frame color compensation parameters;

[0100] The second register group receives the calibration parameters output by the compensation network closed loop for the next frame in the background;

[0101] The two register groups are alternately rotated to realize seamless switching of compensation instruction frames, ensuring no flicker and no lag in high frame rate display environment.

[0102] It also includes redundant wiring and reverse compensation current injection mechanism: when the main driving line detects faults such as resistance abnormality, delay offset, etc., the control module triggers the wiring switching logic to switch the sub-pixel electrode connection to the redundant wiring, and at the same time starts the reverse compensation current injection, the current intensity is defined as follows:

[0103] wherein, represents the compensation current injected in the opposite direction, is the electroluminescent efficiency coefficient of the display panel material (depends on the physical properties of OLED or QLED devices), The rate of change of the current sub-pixel decay gradient over time, The Euclidean distance between the corresponding channel gamut anchor point and the standard color center. This mechanism is used to quickly neutralize the accumulated charge of the walking line, improve the response speed of the pixel, and delay the aging of the electrode material.

[0104] The present application encompasses any alternatives, modifications, equivalent methods and schemes made on the essence and scope of the present application. In order to make the public have a thorough understanding of the present application, specific details are described in the following preferred embodiments of the present application, and the present application can also be fully understood without the description of these details to those skilled in the art. In addition, in order to avoid unnecessary confusion to the essence of the present application, well-known methods, processes, procedures, elements and circuits, etc. are not described in detail.

[0105] The above is only the preferred embodiment of the present application, and it should be pointed out that for ordinary skilled in the art, without departing from the principles of the present application, a number of improvements and refinements can also be made, which should be considered as the protection scope of the present application.

Claims

1. A medical image display system based on color calibration, characterized in that: include: Calibration tensor generation module: This module synchronously acquires the real-time aging fingerprint of the display device, the ambient light interference matrix, and the color gamut feature topology of the image to be displayed to generate a three-dimensional calibration input tensor. The aging fingerprint includes the brightness attenuation gradient distribution of each sub-pixel of the display panel. Closed-loop feedback calibration module: Inputs the three-dimensional calibration input tensor into a pre-trained multimodal adversarial compensation network to generate a pixel-level aging-environment coupling compensation coefficient matrix and color gamut anchor point coordinates. The compensation coefficient matrix and color gamut anchor point coordinates form a closed-loop feedback calibration path; Compensation execution module: Based on the closed-loop feedback calibration path, the input image is subjected to sub-pixel color remapping, and the compensated quantized color instruction set is written into the redundant driving unit of the display panel to achieve the coordinated offset of device aging and environmental interference.

2. The medical image display system based on color calibration according to claim 1, characterized in that: The calibration tensor generation module specifically includes: An aging fingerprint map construction unit detects the luminous efficiency attenuation values ​​of red, green, and blue sub-pixels pixel by pixel in a frame scanning manner through a micro-photoelectric sensor array embedded in the display panel, and constructs the aging fingerprint map based on the spatial distribution of the attenuation values. Ambient light interference matrix generation unit: uses a multi-spectral ambient light probe to synchronously collect the wavelength-intensity interference pattern of ambient light, extracts the coherence characteristics of the light intensity in each band through Fourier transform, and generates an ambient light interference matrix including spectral offset and intensity modulation factor; Color gamut feature topology analysis unit: performs color gamut feature analysis on the image to be displayed, extracts the topological correlation between the chromaticity coordinates and brightness values ​​of each color block in the image in the CIExyY space, and constructs a color gamut feature topology map; A three-dimensional calibration tensor construction unit is configured to perform tensor splicing on the spatial attenuation distribution of the aging fingerprint map, the frequency domain feature vector of the ambient light interference matrix, and the chromaticity correlation weight of the color domain feature topology map to obtain a three-dimensional calibration input tensor.

3. The medical image display system based on color calibration according to claim 2, characterized in that: In the aging fingerprint map, the brightness attenuation gradient distribution of each sub-pixel is expressed as: ； in, Indicates the Sub-pixel type at position The normalized brightness attenuation gradient of Indicates sub-pixel type The initial brightness value at the factory, Indicates the current moment The brightness value of the sub-pixel, Indicates and accumulates working time The related nonlinear aging coefficient, the aging fingerprint is formally expressed as: .

4. The medical image display system based on color calibration according to claim 2, characterized in that: The three-dimensional calibration input tensor is formed along the channel dimension and has a size of H×W×(N+M+P), where H and W are the display panel resolution, N is the number of aging parameter channels, M is the ambient light feature dimension, and P is the color gamut topology association dimension.

5. The medical image display system based on color calibration according to claim 1, characterized in that: The multimodal adversarial compensation network includes a generator branch and a compensation verifier branch, and the closed-loop feedback calibration module specifically includes: Compensation coefficient generation unit: The generator branch uses depthwise separable convolution to extract aging-environment coupling features from the three-dimensional calibration input tensor and generates a pixel-level aging-environment coupling compensation coefficient matrix including red, green, and blue sub-pixel compensation gain values; Color gamut anchor point extraction unit: The compensation verifier branch synchronously processes the color gamut feature topology map in the three-dimensional calibration input tensor, extracts the dynamic distribution characteristics of the image color gamut through an adaptive spatial pooling layer, and outputs a set of color gamut anchor point coordinates in the CIE LAB color space; Generator-verifier adversarial unit: The generator branch and the compensation verifier branch form a dynamic game mechanism through adversarial training: the compensation coefficient matrix is ​​applied to the test image under the simulated aging environment, and the compensated image is input into the compensation verifier branch for color gamut anchor point matching verification. If the color gamut anchor point coordinate offset exceeds the preset threshold, the convolution kernel weight of the generator branch is reversely corrected; Closed-loop calibration path construction unit: This unit performs spatiotemporal correlation analysis between the color gamut anchor point coordinates generated by the current frame and the color gamut feature topology of the next frame, generates a color gamut anchor point drift vector, and feeds it back to the generator branch to adjust the spatial distribution weight of the compensation coefficient matrix in real time. Compensation coefficient nonlinear mapping unit: The pixel-level aging-environment coupling compensation coefficient matrix forms a nonlinear mapping relationship with the color gamut anchor point coordinates.

6. The medical image display system based on color calibration according to claim 5, characterized in that: The adaptive spatial pooling layer extracts the dynamic distribution characteristics of the image color gamut and outputs a color gamut anchor point coordinate set in the CIE LAB color space, specifically including: After receiving the color gamut feature topology map in the calibration input tensor, the compensation verifier branch enhances the key color response areas in the color gamut channel through the channel attention mechanism to obtain the enhanced color gamut feature map. The enhanced color gamut feature map is input into the adaptive spatial pooling layer. The adaptive spatial pooling layer automatically adjusts the pooling window size and sliding step size based on the color distribution density of different regions in the current image, so that the pooling operation adaptively performs fine-grained aggregation on high color variation areas and performs large-scale compression on low-variance areas, preserving the structural information of color transition areas and edges. The compact color gamut representation obtained after pooling is input to the feature compressor, converted to the CIE LAB color space, and clustering and color center extraction are performed on each local area. The representative color center is extracted as the color gamut anchor point, which reflects the main color distribution position of the current frame image in the perceptual space.

7. The medical image display system based on color calibration according to claim 5, characterized in that: The spatiotemporal correlation analysis in the closed-loop calibration path construction unit specifically includes: When performing color gamut anchor point drift analysis, the color gamut anchor point coordinate set extracted from the current frame image is spatially aligned with the color gamut feature topology map in the next frame image to ensure that the corresponding relationship between the color gamut anchor points in the display area between the two frames is consistent. The color distribution characteristics between the anchor points are then used for feature matching to determine the optimal corresponding target for the current frame anchor point in the next frame. For each pair of successfully matched color gamut anchor points, the direction and magnitude of color drift are determined based on their coordinate differences in the CIE LAB space and their position changes in the image space. A cross-frame anchor point drift vector is generated and fed back to the compensation network to adjust the compensation parameter distribution of subsequent frames, thereby maintaining color consistency between consecutive frames.

8. The medical image display system based on color calibration according to claim 5, characterized in that: In the compensation coefficient matrix nonlinear mapping unit, the nonlinear mapping relationship between the pixel-level aging-environment coupling compensation coefficient matrix and the aging fingerprint attenuation gradient and the offset distance between the current color gamut anchor point and the standard color gamut is in the form of a full sigmoid, which is expressed as: ;in, Indicates the The aging fingerprint attenuation gradient of the sub-pixel at the position, Indicates the corresponding color channel The Euclidean distance between the color gamut anchor point and the center of the standard color gamut, Represents the mapping sensitivity factor that is dynamically optimized during adversarial training.

9. The medical image display system based on color calibration according to claim 1, characterized in that: The compensation execution module specifically includes: Brightness correction calculation unit: performs sub-pixel convolution operation on the pixel-level aging-environment coupling compensation coefficient matrix and the original pixel value of the current input image to generate a compensated RGB three-channel brightness correction value; Boundary constraint and coordinate remapping unit: constructs a dynamic color gamut enclosure in the CIE LAB color space according to the color gamut anchor point coordinate set, performs chromaticity boundary constraints on the corrected pixel values, and implements sub-pixel color coordinate remapping through a Lagrange interpolation algorithm; Color instruction writing unit: converts the remapped color data into a color instruction set, which includes the driving pulse width modulation parameters and voltage gradient compensation values ​​of each sub-pixel, and is written in real time through the spare TFT circuit in the redundant driving unit built into the display panel.

10. The medical image display system based on color calibration according to claim 1, characterized in that: The compensation execution module is further provided with a double buffer register, specifically including providing a double buffer register group in the redundant drive unit; When the first register group executes the current frame compensation instruction, the second register group synchronously receives the closed-loop feedback calibration parameters of the next frame, realizing low-latency switching between aging compensation and environmental interference cancellation.

Images