Display dynamic partition image rendering method and system based on content identification

Abstract

The invention discloses a display dynamic partition image rendering method and system based on content recognition, and the method comprises the steps: firstly carrying out the classification of content types through employing a built-in RGB data detection module of a display, combining global recognition with an 8 * 6 grid partition recognition mechanism, and employing a lightweight CNN model and a long-short-term memory network; and an adaptive rendering engine dynamically adjusts a rendering strategy, a transition processing mechanism ensures smooth parameter transition during content switching through a time delay gradual change algorithm, and cooperative work of super-resolution sharpening of a text region and wide color gamut mapping of an image region is realized in cooperation with an edge detection technology. The system is also integrated with a real-time feedback learning module which allows a user to automatically construct a multi-dimensional preference database containing RGB data, content types and rendering parameters after marking unsatisfied areas. The problems of color distortion, insufficient layering sense, complicated operation and the like caused by a fixed rendering mode of a traditional display are effectively solved, and the picture display effect and the user watching experience are remarkably improved.

Description

Technical Field

[0001] This invention relates to the field of display applications, and in particular to a method and system for dynamic partitioned image rendering of displays based on content recognition. Background Technology

[0002] With the increasing richness and diversity of multimedia content, users' demands for monitor image quality are also constantly rising. Traditional monitor rendering technology often uses a fixed rendering mode, which cannot dynamically adjust according to the actual needs of the content, resulting in suboptimal image quality. Users often need to manually switch rendering modes based on the content, which is not only cumbersome but also makes it difficult to ensure that the image remains at its best throughout.

[0003] Common monitors typically employ fixed rendering modes (such as text mode / cinema mode), requiring manual switching by the user and failing to automatically optimize based on the content. Especially in mixed content scenarios (such as web pages with mixed text and images), a single rendering mode results in blurry or unclear text, color distortion, and a lack of depth, failing to meet the requirements of display applications. To address this, a content recognition-based dynamic partitioned image rendering method and system for monitors is proposed. Summary of the Invention

[0004] This invention provides the following technical solution: a content-recognition-based dynamic partitioned image rendering method for displays, comprising the following steps:

[0005] S1 Dynamic Content Recognition:

[0006] By using the monitor's built-in image RGB data detection and analysis module, the image features of the displayed screen are analyzed in real time. At the same time, a global recognition mode is used to analyze the image features of the central area of ​​the displayed screen in real time, and a partition recognition mode is used to analyze the image features of the 8×6 grid partitions of the displayed screen in real time. Furthermore, a lightweight CNN model and a temporal analysis model of 5 consecutive frames are used to classify the content types, which include text, images, and videos. By analyzing screen features and classifying content types in real time, a foundation is provided for accurate rendering, thereby improving the screen display effect.

[0007] S2 Adaptive Rendering Engine:

[0008] Based on the content type identification results in step S1, the rendering strategy is dynamically adjusted. In text mode, sharpness is maximized, and in image mode, color gamut, saturation, and dynamic range are maximized. At the same time, a dynamic color space mapping algorithm is adopted. When the input signal is sRGB, it is automatically converted to DCI-P3 color gamut while maintaining color reproduction error. When HDR10+ metadata is detected, 12-bit color depth rendering is enabled to improve the smoothness of color transition between bright and dark fields. For mixed content scenes, independent mapping of partitioned color spaces is adopted. That is, the text area maintains the sRGB color gamut to ensure readability, while the video area is mapped to a wide color gamut and the dynamic tone mapping algorithm is activated. By dynamically adjusting the rendering strategy, the picture quality is optimized, making the picture clearer, the colors richer, and improving the viewing experience.

[0009] S3 transition handling mechanism:

[0010] When switching content types in step S2, a time-delay gradient algorithm is used. In mixed content scenarios, rendering is performed by region segmentation, with text regions being sharpened independently and video and image regions being rendered independently. At the edges of the partitions, a blur algorithm is used to gradually weaken the color and then turn off the color enhancement, so as to achieve a smooth transition of the image, avoid abrupt changes, and improve the viewing comfort in mixed content scenarios.

[0011] S4 Real-time Quality Feedback and Self-Learning:

[0012] Users mark unsatisfactory areas through the OSD menu. The system then automatically collects the RGB data, content type, and rendering parameters of that area to build a user preference database. Using reinforcement learning algorithms, the system dynamically adjusts the weights of rendering parameters based on historical preference data and optimizes the rendering effect based on user feedback to meet personalized needs and improve user satisfaction.

[0013] This invention provides a content-recognition-based dynamic partitioned image rendering system for displays, employing the aforementioned content-recognition-based dynamic partitioned image rendering method for displays, including:

[0014] The system includes an image detection module, an image analysis module, an image output module, an adaptive rendering engine, a transition processing module, and a real-time feedback learning module. The image detection module is connected to the image analysis module via a data bus. The image detection module is used to capture 8×6 grid partition data of the display screen through the image sensor built into the display.

[0015] The image analysis module is used to receive RGB data from the image detection module. The image analysis module is internally equipped with a lightweight CNN model and a long short-term memory network temporal analysis model. The image analysis module is connected to the adaptive rendering engine through control signal lines.

[0016] An adaptive rendering engine is used to dynamically adjust rendering parameters based on the content type recognition results of the image analysis module. The adaptive rendering engine is connected to the transition processing module via a data link. The transition processing module is used to achieve a smooth transition of screen content by using a time-delay gradient algorithm when the adaptive rendering engine module switches rendering strategies.

[0017] The real-time feedback learning module is used to receive the coordinates of the unsatisfactory areas marked by the user through the OSD menu, and to collect the RGB data, content type and rendering parameters of the unsatisfactory areas and store them in the user preference database. The real-time feedback learning module is connected to the adaptive rendering engine through a bidirectional data interface.

[0018] Preferably, the specific steps for text recognition of content type in step S1 are as follows:

[0019] First, the Sobel gradient features of the character edges are extracted. When the mean gradient value is greater than the threshold T1, it indicates that the edge transition is obvious.

[0020] Secondly, the dispersion of the image colors is calculated using the color histogram entropy method. This involves dividing the image into 16 intervals based on the RGB grayscale values, from smallest to largest, to form a color histogram. The number of pixels in each of these 16 intervals is then counted, and the probability of each interval's pixel count relative to the total number of pixels is calculated. The entropy value is also calculated. When the entropy value is less than a threshold T2, it indicates that the color distribution is concentrated and the dispersion is low. When both clear edge transitions and concentrated color distribution are met, the image is identified as text. By accurately identifying text through edge features and color distribution, text clarity and readability are improved.

[0021] Preferably, in step S1, when classifying content types using a lightweight CNN model, the lightweight CNN model employs a depthwise separable convolutional structure. For text types, it focuses on extracting high-frequency edge features and low color variance features; for image and video types, it focuses on extracting continuous tone variations and high color depth features; and it optimizes feature extraction for content types, thereby improving recognition accuracy and efficiency and providing support for accurate rendering.

[0022] Preferably, the continuous 5-frame image temporal analysis model in step S1 adopts a long short-term memory network structure. By analyzing the content type change trend of historical frames, the content type of the next frame is predicted probabilistically. When the prediction confidence of the corresponding type reaches a preset high threshold, the loading of the corresponding rendering strategy is triggered in advance. Predicting the content type change and triggering the rendering strategy in advance improves the smoothness of the screen and the response speed.

[0023] Preferably, in step S2, the dynamic color space mapping algorithm, when converting the sRGB signal to the DCI-P3 color gamut, establishes a color mapping lookup table and combines it with a color gamut compression algorithm to ensure that the converted colors remain consistent within the range of human visual perception. Furthermore, the 12-bit color depth rendering enabled when HDR10+ metadata is detected increases the color quantization level, making the color transition between the highlights of sunlight halos and night scene lights and the dark scenes in the image smoother and more natural, eliminating the color banding phenomenon that occurs under 8-bit color depth, accurately controlling color performance, improving the color richness and transition smoothness of the image, and enhancing the visual effect.

[0024] Preferably, in step S3, the time-delay gradient algorithm uses linear interpolation to smoothly transition the rendering parameters of brightness, contrast, and color when switching content types. The transition duration is automatically adjusted according to the content switching type. Furthermore, in mixed content scenarios, the rendering is performed by region segmentation. The outline of the text region is located through the edge detection algorithm. The super-resolution sharpening algorithm is enabled separately for the text region, while wide color gamut mapping and dynamic contrast optimization are enabled separately for the video and image regions. This achieves smooth switching of rendering strategies, avoids sudden changes in the screen, and improves viewing comfort.

[0025] Preferably, in step S4, when the user marks an unsatisfactory area through the OSD menu, the system synchronously records the coordinate position of the marked area, the frame information of the screen content, and the current combination of rendering parameters to establish a multi-dimensional database containing the user's subjective preferences. At the same time, the reinforcement learning algorithm constructs a state, action, and reward model based on the user preference database, dynamically adjusts the rendering parameter weights under different content types, and dynamically adjusts the rendering parameters according to the user's preferences to achieve personalized display effects and improve user satisfaction.

[0026] Preferably, the image detection module includes a global recognition unit and a partition recognition unit. The global recognition unit is used to analyze the image features of the central area of ​​the displayed screen in real time. The partition recognition unit is used to divide the screen into 8×6 grid partitions, collect image data for each of the 48 independent partitions, and output the content type parameters corresponding to each partition. The lightweight CNN model in the image analysis module adopts a hierarchical feature extraction structure. The bottom layer network of the hierarchical feature extraction structure extracts high-frequency edge features of the text through depthwise separable convolution. When obvious character edge transitions are detected, an edge enhancement algorithm is triggered. The high layer network of the hierarchical feature extraction structure is used to analyze the continuous tone change features of the image and judge the image and video content through color distribution statistics. Through global and partition recognition, the screen content information is accurately obtained, providing reliable data for subsequent processing.

[0027] Preferably, the adaptive rendering engine includes a color space conversion unit and a partition rendering control unit. When the input signal is sRGB, the color space conversion unit converts the color gamut to a wide color gamut space through a built-in color mapping lookup table. When HDR1+ metadata is detected, the high color depth rendering mode is automatically enabled. The partition rendering control unit is used to apply independent color gamut mapping strategies to text areas and image areas respectively for mixed content scenes, flexibly converting color gamut and controlling rendering, improving the color performance and contrast of the picture, and optimizing the visual effect.

[0028] In summary, compared with the prior art, the present invention provides a method and system for dynamic partitioning image rendering of a display based on content recognition, which has the following beneficial effects:

[0029] 1. This invention, through its built-in image RGB data detection and analysis module, enables real-time analysis of the image features of the displayed screen. This allows it to move beyond fixed rendering modes and dynamically identify screen content, including various types such as text, images, and videos. This enables the monitor to automatically adjust to the best or most suitable display effect according to the actual needs of the screen content, without requiring manual switching by the user, greatly improving the user experience. Furthermore, in mixed content scenarios, through the partition recognition mode and adaptive rendering engine, it can independently adjust the rendering strategy for different content types in different areas, effectively solving the problems of blurry or unclear text or color distortion caused by traditional single rendering modes, resulting in a stronger sense of image layering and richer details.

[0030] 2. This invention achieves precise control over screen colors through a dynamic color space mapping algorithm and independent color space mapping technology. When the input signal is sRGB, it automatically converts to the DCI-P3 color gamut while maintaining accurate color reproduction. When HDR10+ metadata is detected, 12-bit color depth rendering is enabled, improving the smoothness of color transitions between bright and dark areas. This not only optimizes the color performance of the screen but also allows the display to automatically select the most suitable color space mapping strategy when presenting different content. Furthermore, by maintaining the sRGB color gamut of text areas to ensure readability, and mapping video areas to a wide color gamut and activating a dynamic tone mapping algorithm, the overall color performance and sense of depth of the screen are further enhanced. This solves the problems of color distortion and weak sense of depth caused by a single rendering mode in existing technologies. Attached Figure Description

[0031] Figure 1 This is a flowchart of the method of the present invention.

[0032] Figure 2 This is a system architecture block diagram of the present invention. Detailed Implementation

[0033] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0034] Please see Figure 1 This invention provides a technical solution for a content-recognition-based dynamic partitioned image rendering method for displays, comprising the following steps:

[0035] S1 Dynamic Content Recognition:

[0036] The system utilizes the built-in RGB data detection and analysis module to analyze the image features of the displayed screen in real time. It employs a global recognition mode to analyze the image features of the central area of ​​the screen in real time, and a partitioned recognition mode to analyze the image features of the 8×6 grid partitions of the screen in real time. Furthermore, it classifies content types, including text, images, and videos, using a lightweight CNN model and a temporal analysis model of five consecutive frames. The lightweight CNN model employs a depthwise separable convolutional structure, focusing on extracting high-frequency edge features and low color variance features for text, and continuous tone variations and high color depth features for images and videos. The temporal analysis model of five consecutive frames uses a long short-term memory network structure. By analyzing the content type change trends of historical frames, it predicts the content type of the next frame probabilistically. When the prediction confidence of the corresponding type reaches a preset high threshold, the corresponding rendering strategy is pre-triggered.

[0037] The specific steps for text recognition of content types are as follows:

[0038] First, the Sobel gradient features of the character edges are extracted. When the mean gradient value is greater than the threshold T1, it indicates that the edge transition is obvious.

[0039] Secondly, the dispersion of the image color is calculated using the color histogram entropy method. That is, based on the size of the RGB grayscale values, 16 intervals are divided from small to large as the color histogram, and the number of pixels in each of the 16 intervals is counted. Then, the probability value of the number of pixels in the 16 intervals relative to the total number of pixels is calculated, and the entropy value is calculated. When the entropy value is less than the threshold T2, it indicates that the color distribution is concentrated and the dispersion is low. When both the obvious edge transition and the concentrated color distribution conditions are met, it is judged as text.

[0040] S2 Adaptive Rendering Engine:

[0041] Based on the content type identification results in step S1, the rendering strategy is dynamically adjusted. In text mode, sharpness is maximized, and in image mode, color gamut, saturation, and dynamic range are maximized. At the same time, a dynamic color space mapping algorithm is adopted. When the input signal is sRGB, it is automatically converted to DCI-P3 color gamut while maintaining color reproduction error. When HDR10+ metadata is detected, 12-bit color depth rendering is enabled to improve the smoothness of color transition between bright and dark fields. For mixed content scenes, independent mapping of partitioned color spaces is adopted. That is, the text area maintains the sRGB color gamut to ensure readability, while the video area is mapped to a wide color gamut and the dynamic tone mapping algorithm is activated.

[0042] When converting sRGB signals to the DCI-P3 color gamut, the dynamic color space mapping algorithm establishes a color mapping lookup table and combines it with a color gamut compression algorithm to ensure that the converted colors remain consistent within the range of human visual perception. Furthermore, the 12-bit color depth rendering enabled when HDR10+ metadata is detected increases the number of color quantization levels, making the color transition between the highlights of sunlight halos and night scene lights and the dark scenes in the image smoother and more natural, and eliminating the color banding phenomenon that occurs under 8-bit color depth. The specific implementation process of the above method is as follows:

[0043] When the algorithm receives the input sRGB signal, it first initiates the color mapping process. The core of this process lies in establishing a color mapping lookup table. This lookup table is not generated out of thin air, but is built upon a deep understanding and precise correspondence between the sRGB color gamut and the target DCI-P3 color gamut. The algorithm analyzes the characteristics of various colors in the sRGB color gamut, including their hue, saturation, and brightness. Then, based on the wider color range of the DCI-P3 color gamut, it finds the closest corresponding color in the DCI-P3 color gamut that is consistent within the range of human visual perception for each color in the sRGB color gamut. This correspondence is not a simple linear mapping, but is carefully designed and optimized to ensure that the converted colors appear visually natural and consistent.

[0044] While building the color mapping lookup table, the algorithm also incorporates a color gamut compression algorithm. The purpose of this algorithm is to intelligently compress certain colors in the sRGB color space into the DCI-P3 color gamut when those colors exceed the expressive power of the DCI-P3 gamut, while preserving as much of the original color's main characteristics and visual effect as possible. This compression is not a simple cropping or scaling process, but rather a series of complex calculations and optimizations to ensure that the compressed colors are reasonably distributed in the color space and that the difference from human visual perception is minimized.

[0045] By combining a color mapping lookup table and a color gamut compression algorithm, the algorithm ensures that when converting sRGB signals to the DCI-P3 color gamut, the converted colors remain highly consistent within the range of human visual perception. This means that even if the original signal comes from the relatively narrow sRGB color space, the image processed by the algorithm can present richer and more vibrant colors, while avoiding color distortion or deviation caused by color gamut mismatch.

[0046] When the algorithm detects that the input signal contains HDR10+ metadata, it further enables 12-bit color depth rendering. Compared to the traditional 8-bit color depth, 12-bit color depth can provide more levels of color quantization. This means that when representing the same color, 12-bit color depth can more finely divide the brightness levels of the color, thus presenting a smoother and more natural color transition;

[0047] When 12-bit color depth rendering is enabled, the algorithm performs special optimizations for both bright and dark scenes. For example, in bright scenes such as sunlight halos or nighttime lights, the algorithm ensures smoother color transitions in these areas, avoiding banding caused by insufficient color quantization levels. Similarly, in dark scenes, the algorithm ensures better preservation of color gradation and detail, making the image look more natural and realistic.

[0048] By increasing the number of color quantization levels and optimizing color transitions between bright and dark scenes, 12-bit color depth rendering significantly enhances the overall color performance and visual effects of the image. It not only eliminates the color banding phenomenon commonly seen in 8-bit color depth rendering, but also makes the colors in the image more delicate and rich, providing viewers with a more immersive viewing experience.

[0049] S3 transition handling mechanism:

[0050] When switching content types in step S2, a time-delay gradient algorithm is used. In mixed content scenarios, rendering is performed by region segmentation, with text regions being sharpened independently and video and image regions being rendered independently. At the edges of the partitions, a blur algorithm is used to gradually weaken the color and transition to turning off color enhancement.

[0051] When switching content types, the time-delay gradient algorithm uses linear interpolation to smoothly transition the rendering parameters of brightness, contrast and color, and the transition time is automatically adjusted according to the content switching type. In mixed content scenes, it performs region segmentation rendering, locates the outline of the text region through the edge detection algorithm, enables super-resolution sharpening algorithm separately for the text region, and enables wide color gamut mapping and dynamic contrast optimization separately for video and image regions. The specific implementation process of the above method is as follows:

[0052] When a content type switch is detected, the time-delayed fading algorithm immediately initiates a smooth transition process. The core of this process lies in using linear interpolation to progressively adjust key rendering parameters such as brightness, contrast, and color. Linear interpolation is a simple and effective mathematical method that can uniformly generate a series of intermediate values ​​between two known values. In the time-delayed fading algorithm, linear interpolation is used to ensure that rendering parameters smoothly transition from values ​​corresponding to one content type to values ​​corresponding to another.

[0053] At the moment of content type switching, the algorithm obtains the rendering parameter values ​​before and after the switch. Then, based on the preset transition duration, the algorithm calculates the intermediate parameter values ​​that should be applied at each time point. These intermediate parameter values ​​are calculated through linear interpolation, which ensures that the parameter changes are continuous and smooth, thereby avoiding jumps or flickering caused by sudden parameter changes in the image;

[0054] It's worth noting that the transition duration is not fixed but automatically adjusted based on the content switching type. For example, when switching from text mode to image mode, because the two modes differ significantly in rendering parameters, the algorithm may set a relatively long transition duration to ensure the image gradually adapts to the change. Conversely, when switching from image mode to video mode, because the two modes may have more similar rendering parameters, the algorithm may set a shorter transition duration to reduce unnecessary waiting time.

[0055] In mixed content scenarios, the time-delay gradient algorithm is further combined with region-based segmentation rendering technology to achieve more refined image optimization. The core of region-based segmentation rendering technology lies in accurately identifying and locating the outlines of text regions in the image through edge detection algorithms. Edge detection algorithms can analyze changes in the brightness, color, and other attributes of pixels in the image, thereby identifying the boundaries between different regions. In mixed content scenarios, text regions typically have obvious edge features, so edge detection algorithms can effectively distinguish them from other regions (such as video or image regions).

[0056] Once the outline of the text region is accurately identified, the time-delay gradient algorithm will apply a separate super-resolution sharpening algorithm to that region. Super-resolution sharpening enhances the contrast and clarity of text edges, making the text stand out more and be easier to read. Meanwhile, for video and image regions, the algorithm will apply wide color gamut mapping and dynamic contrast optimization techniques separately. Wide color gamut mapping expands the color range of the image, resulting in richer and more vibrant colors in videos and images; while dynamic contrast optimization automatically adjusts contrast parameters based on changes in the brightness of the content, better preserving details and depth in the image.

[0057] S4 Real-time Quality Feedback and Self-Learning:

[0058] Users mark unsatisfactory areas through the OSD menu. Subsequently, the system automatically collects the RGB data, content type, and rendering parameters of that area to establish a user preference database. Then, a reinforcement learning algorithm is used to dynamically adjust the weights of rendering parameters based on historical preference data. When users mark unsatisfactory areas through the OSD menu, the system simultaneously records the coordinate position of the marked area, the frame information of the screen content, and the current combination of rendering parameters to establish a multi-dimensional database containing the user's subjective preferences. At the same time, the reinforcement learning algorithm constructs a state, action, and reward model based on the user preference database and dynamically adjusts the weights of rendering parameters under different content types. The specific implementation process of the above method is as follows:

[0059] When users view the monitor screen, if they find that certain areas do not display as expected, they can mark them using the monitor's OSD menu. The OSD menu is a user interface provided by the monitor, allowing users to interact with the monitor through simple button operations. To mark an unsatisfactory area, users simply use the arrow keys to move the cursor to the target area and then press the confirmation button to complete the marking.

[0060] Once a user marks an area of ​​dissatisfaction, the system responds immediately. First, the system records the coordinates of the marked area for precise analysis later. Simultaneously, it collects the RGB data of the area, reflecting its actual color representation in the image. Furthermore, the system obtains the current content type and rendering parameter combination to understand under what rendering strategy the area is displaying the unsatisfactory effect.

[0061] The system stores the collected data in a user preference database. This database is specifically designed to store users' subjective preferences and feedback, including their evaluations and expectations of the visual effects in different scenarios. By continuously accumulating user preference data, the system can gradually understand users' personalized needs, providing strong support for subsequent optimization of rendering effects.

[0062] To dynamically adjust rendering parameter weights, the system employs a reinforcement learning algorithm. Reinforcement learning is a machine learning method that allows a system to learn optimal strategies through interaction with its environment. In this invention, the reinforcement learning algorithm is used to construct state, action, and reward models. The state model describes the combination of the current screen content and rendering parameters, the action model defines the rendering parameter adjustment strategies the system can adopt, and the reward model evaluates the effectiveness of each action based on user feedback.

[0063] During reinforcement learning, the system continuously tries different combinations of rendering parameters and evaluates their effectiveness based on user feedback. When a user marks an area as unsatisfactory, the system considers the current rendering parameter combination to be performing poorly in that area and therefore awards a negative reward. Conversely, when a user does not mark any areas or marks a satisfactory area, the system considers the current rendering parameter combination to be performing well and therefore awards a positive reward.

[0064] Through continuous iteration and optimization, reinforcement learning algorithms can gradually learn the optimal rendering parameter weights for different content types. These weights reflect users' expectations and preferences for visual effects in different scenarios, and they are used to dynamically adjust rendering strategies to ensure that the image always maintains the display effect expected by the user.

[0065] Please see Figure 2 The present invention provides a content-recognition-based dynamic partitioned image rendering system for displays, employing the aforementioned content-recognition-based dynamic partitioned image rendering method for displays, comprising:

[0066] The system includes an image detection module, an image analysis module, an image output module, an adaptive rendering engine, a transition processing module, and a real-time feedback learning module. The image detection module is connected to the image analysis module via a data bus. The image detection module is used to capture 8×6 grid partition data of the display screen through the image sensor built into the monitor.

[0067] The image analysis module receives RGB data from the image detection module. Internally, the image analysis module contains a lightweight CNN model and a long short-term memory network temporal analysis model. The image analysis module is connected to the adaptive rendering engine via control signal lines.

[0068] The adaptive rendering engine dynamically adjusts rendering parameters based on the content type recognition results of the image analysis module. The adaptive rendering engine is connected to the transition processing module via a data link. The transition processing module uses a time-delay gradient algorithm to achieve a smooth transition of screen content when the adaptive rendering engine module switches rendering strategies.

[0069] The real-time feedback learning module is used to receive the coordinates of the unsatisfactory areas marked by the user through the OSD menu, and to collect the RGB data, content type and rendering parameters of the unsatisfactory areas and store them in the user preference database. The real-time feedback learning module is connected to the adaptive rendering engine through a bidirectional data interface.

[0070] The image detection module includes a global recognition unit and a partition recognition unit. The global recognition unit is used to analyze the image features of the central area of ​​the display screen in real time. The partition recognition unit is used to divide the screen into 8×6 grid partitions, collect image data for each of the 48 independent partitions, and output the content type parameters corresponding to each partition. The lightweight CNN model in the image analysis module adopts a hierarchical feature extraction structure. The bottom layer of the hierarchical feature extraction structure extracts the high-frequency edge features of the text through depthwise separable convolution. When a clear transition of character edges is detected, the edge enhancement algorithm is triggered. The high layer of the hierarchical feature extraction structure is used to analyze the continuous tonal change features of the image and judge the image and video content through color distribution statistics.

[0071] The adaptive rendering engine includes a color space conversion unit and a partition rendering control unit. When the input signal is sRGB, the color space conversion unit converts the color gamut to a wide color gamut space through a built-in color mapping lookup table. When HDR1+ metadata is detected, the high color depth rendering mode is automatically enabled. The partition rendering control unit is used to apply independent color gamut mapping strategies to text areas and image areas respectively for mixed content scenes.

[0072] This solution utilizes a built-in image RGB data detection and analysis module to achieve real-time analysis of the image features of the displayed screen. This allows it to move beyond fixed rendering modes and dynamically identify screen content, including various types such as text, images, and videos. The monitor can then automatically adjust to the optimal or most suitable display effect based on the actual needs of the screen content, eliminating the need for manual switching and significantly improving the user experience. Furthermore, in mixed content scenarios, through a partitioned recognition mode and an adaptive rendering engine, it can independently adjust the rendering strategy for different content types in different areas. This effectively solves the problems of blurry or unclear text or color distortion caused by traditional single rendering modes, resulting in a stronger sense of depth and richer details in the image.

[0073] This solution achieves precise control over screen colors through a dynamic color space mapping algorithm and independent color space mapping technology. When the input signal is sRGB, it automatically converts to the DCI-P3 color gamut while maintaining accurate color reproduction. When HDR10+ metadata is detected, 12-bit color depth rendering is enabled, improving the smoothness of color transitions between bright and dark areas. This not only optimizes the screen's color performance but also allows the monitor to automatically select the most suitable color space mapping strategy when displaying different content. Furthermore, by maintaining the sRGB color gamut for text areas to ensure readability, and mapping video areas to a wide color gamut and activating a dynamic tone mapping algorithm, the overall color performance and depth of the screen are further enhanced. This solves the problems of color distortion and weak depth caused by a single rendering mode in existing technologies.

[0074] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0075] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A content-recognition-based method for dynamic partitioning image rendering on a display, characterized in that, include: S1 Dynamic Content Recognition: The display uses a built-in RGB data detection and analysis module to analyze the image features of the displayed screen in real time. It also uses a global recognition mode to analyze the image features of the central area of ​​the displayed screen in real time, and a partition recognition mode to analyze the image features of the 8×6 grid partitions of the displayed screen in real time. Furthermore, it uses a lightweight CNN model and a time-series analysis model of 5 consecutive frames to classify the content type. S2 Adaptive Rendering Engine: Based on the content type identification results in step S1, the rendering strategy is dynamically adjusted. At the same time, a dynamic color space mapping algorithm is adopted. When the input signal is sRGB, it is automatically converted to DCI-P3 color gamut while maintaining color reproduction error. When HDR10+ metadata is detected, 12-bit color depth rendering is enabled to improve the smoothness of color transition between bright and dark fields. For mixed content scenes, independent mapping of partitioned color spaces is adopted. S3 transition handling mechanism: When switching content types in step S2, a time-delay gradient algorithm is used. In mixed content scenarios, rendering is performed by region segmentation, with text regions being sharpened independently and video and image regions being rendered independently. At the edges of the partitions, a blur algorithm is used to gradually weaken the color and transition to turning off color enhancement. S4 Real-time Quality Feedback and Self-Learning: Users mark unsatisfactory areas through the OSD menu. The system then automatically collects the RGB data, content type, and rendering parameters of the area to build a user preference database. It then uses a reinforcement learning algorithm to dynamically adjust the weights of the rendering parameters based on historical preference data.

2. The content-recognition-based dynamic partitioning image rendering method for displays according to claim 1, characterized in that: The specific steps for text recognition of content types in step S1 are as follows: First, the Sobel gradient features of the character edges are extracted. When the mean gradient value is greater than the threshold T1, it indicates that the edge transition is obvious. Secondly, the dispersion of the image colors is calculated using the color histogram entropy method. That is, based on the RGB grayscale values, the image is divided into 16 intervals from smallest to largest as a color histogram, and the number of pixels in each of the 16 intervals is counted. Then, the probability value of the number of pixels in the 16 intervals relative to the total number of pixels is calculated, and the entropy value is calculated. When the entropy value is less than the threshold T2, it indicates that the color distribution is concentrated and the dispersion is low. When both the obvious edge transition and the concentrated color distribution conditions are met, it is determined to be text.

3. The content-recognition-based dynamic partitioned image rendering method for displays according to claim 1, characterized in that: In step S1, when classifying content types using a lightweight CNN model, the lightweight CNN model employs a depthwise separable convolutional structure and focuses on extracting high-frequency edge features and low color variance features for text types, while focusing on extracting continuous tone variations and high color depth features for image and video types.

4. The content-recognition-based dynamic partitioned image rendering method for displays according to claim 1, characterized in that: The five consecutive frames of image temporal analysis model in step S1 adopts a long short-term memory network structure. By analyzing the content type change trend of historical frames, it makes a probability prediction of the content type of the next frame. When the confidence of the corresponding type prediction reaches a preset high threshold, the loading of the corresponding rendering strategy is triggered in advance.

5. The content-recognition-based dynamic partitioning image rendering method for displays according to claim 1, characterized in that: The dynamic color space mapping algorithm in step S2 converts the sRGB signal to the DCI-P3 color gamut by establishing a color mapping lookup table and combining it with a color gamut compression algorithm. This ensures that the converted colors remain consistent within the range of human visual perception. Furthermore, the 12-bit color depth rendering enabled when HDR10+ metadata is detected increases the color quantization level, making the color transition between the highlights of sunlight halos and night scene lights and the colors of dark scenes smoother and more natural, thus eliminating the color banding phenomenon that occurs under 8-bit color depth.

6. The content-recognition-based dynamic partitioned image rendering method for displays according to claim 1, characterized in that: In step S3, the time-delay gradient algorithm uses linear interpolation to smoothly transition the rendering parameters of brightness, contrast, and color when switching content types. The transition duration is automatically adjusted according to the content switching type. In mixed content scenes, it performs region-based rendering, locates the outline of the text region through the edge detection algorithm, enables super-resolution sharpening algorithm separately for the text region, and enables wide color gamut mapping and dynamic contrast optimization separately for the video and image regions.

7. The content-recognition-based dynamic partitioned image rendering method for displays according to claim 1, characterized in that: In step S4, when the user marks an unsatisfactory area through the OSD menu, the system synchronously records the coordinate position of the marked area, the frame information of the screen content, and the current combination of rendering parameters to establish a multi-dimensional database containing the user's subjective preferences. At the same time, the reinforcement learning algorithm constructs a state, action, and reward model based on the user preference database and dynamically adjusts the rendering parameter weights under different content types.

8. A content-recognition-based dynamic partitioned image rendering system for displays, employing the content-recognition-based dynamic partitioned image rendering method for displays according to any one of claims 1-7, characterized in that, include: An image detection module is connected to an image analysis module via a data bus. The image detection module is used to capture 8×6 grid partition data of the display screen through the image sensor built into the display. The image analysis module is used to receive RGB data from the image detection module. The image analysis module is internally equipped with a lightweight CNN model and a long short-term memory network temporal analysis model. The image analysis module is connected to the adaptive rendering engine through control signal lines. An adaptive rendering engine is used to dynamically adjust rendering parameters based on the content type recognition results of the image analysis module. The adaptive rendering engine is connected to the transition processing module via a data link. The transition processing module is used to achieve a smooth transition of screen content by employing a time-delay gradation algorithm when the adaptive rendering engine module switches rendering strategies. The real-time feedback learning module is used to receive the coordinates of the unsatisfactory areas marked by the user through the OSD menu, and to collect the RGB data, content type and rendering parameters of the unsatisfactory areas and store them in the user preference database. The real-time feedback learning module is connected to the adaptive rendering engine through a bidirectional data interface.

9. The content-recognition-based display dynamic partitioning image rendering system according to claim 8, characterized in that: The image detection module includes a global recognition unit and a partition recognition unit. The global recognition unit is used to analyze the image features of the central area of ​​the display screen in real time. The partition recognition unit is used to divide the screen into 8×6 grid partitions, collect image data for each of the 48 independent partitions, and output the content type parameters corresponding to each partition. The lightweight CNN model in the image analysis module adopts a hierarchical feature extraction structure. The bottom layer network of the hierarchical feature extraction structure extracts high-frequency edge features of the text through depthwise separable convolution. When obvious character edge transitions are detected, an edge enhancement algorithm is triggered. The high layer network of the hierarchical feature extraction structure is used to analyze the continuous tonal variation features of the image and to judge the image and video content through color distribution statistics.

10. The content-recognition-based dynamic partitioned image rendering system for displays according to claim 8, characterized in that: The adaptive rendering engine includes a color space conversion unit and a partition rendering control unit. When the input signal is sRGB, the color space conversion unit converts the color gamut to a wide color gamut space through a built-in color mapping lookup table. When HDR1+ metadata is detected, the high color depth rendering mode is automatically enabled. The partition rendering control unit is used to apply independent color gamut mapping strategies to text areas and image areas respectively for mixed content scenes.

Images