CN-116403537-B - Liquid crystal display area dynamic dimming method based on image saliency ranking model

Abstract

The invention discloses a liquid crystal display area dynamic dimming method based on an image saliency ranking model, which comprises the following steps of 1, obtaining an initial backlight value of each backlight partition of an input image, predicting saliency scores of different saliency objects in the input image by adopting an example level saliency ranking detection model based on graph reasoning, performing saliency ranking to obtain a saliency ranking map, 2, dividing the saliency ranking map obtained in the step 1 into a plurality of backlight partitions according to the partition number of backlight sources, giving different weights to the backlight partition where each saliency object is located according to the saliency ranking, then carrying out weighted calculation and correction on the initial backlight value according to the weights to obtain a final partition backlight value, and 3, simulating distribution of the partition backlight value on a liquid crystal panel by adopting a BMA (body-building block) mixed light diffusion algorithm, determining a compensation value of pixels, and displaying the input image by the backlight value and the pixel compensation value. The invention can improve the display effect perceived by human eyes and reduce energy consumption.

Inventors

• QIU LONGZHEN
• NIE XIANHUI
• LIU XIN
• LI HAO
• FANG YONG

Assignees

• 合肥工业大学

Dates

Publication Date

20260512

Application Date

20230407

Claims (2)

1. 1. The liquid crystal display area dynamic dimming method based on the image saliency ranking model is characterized by comprising the following steps of: step 1, converting an input image into a gray level image, partitioning the gray level image according to the partition number of backlight sources, extracting partition backlight values of each backlight partition by adopting different methods, and calculating initial backlight values of each backlight partition based on the partition backlight values; The method comprises the steps of preprocessing an input image, adopting an example level saliency ranking detection model based on graph reasoning, firstly adjusting the size of the input image, then predicting saliency scores of different saliency objects in the input image to perform saliency ranking, and finally restoring the size of the output image to obtain a final saliency ranking graph; Step 2, dividing the backlight partition obtained in the step 1 into blocks corresponding to the backlight partition according to the backlight partition determined in the step 1, giving different weights to the blocks where each salient object is located according to the salient ranks corresponding to each salient object obtained in the step 1, and carrying out weighted calculation correction on the initial backlight value of each block according to the weights, so as to determine the final block backlight value of each block; step 3, simulating the distribution of the block backlight values of each block obtained in the step 2 on the liquid crystal panel by adopting a BMA (light mixing and diffusion) algorithm, determining the compensation value of the pixel, and displaying an input image according to the block backlight value of each block and the compensation value of the pixel; in step 1, a sub-pixel maximum value method is adopted, the value of the channel with the largest brightness information in R, G, B channels of each pixel point in an input image is firstly used as the brightness value of the pixel point, then the maximum value of the brightness values of the pixel points in the subareas is used as a first subarea backlight value BL1, an average value method is adopted, gray values of all the pixel points in each subarea are accumulated, then the average value is taken as a second subarea backlight value BL2, an initial backlight value BL3 of each backlight subarea is calculated based on the first subarea backlight value BL1 and the second subarea backlight value BL2, and the initial backlight BL3 has the following calculation formula: ; In the step 1, an example level saliency ranking detection model based on graph reasoning refers to an improved Mask R-CNN network to carry out saliency example segmentation, a saliency sequencing branch is added to sequence the relative saliency of each segmentation example, different saliency objects in an input image are sequenced according to the importance degree of the saliency objects in the image, namely the time period of simulating the staring of human eyes on different contents in the image, the saliency sequencing is carried out, and backlight extraction is carried out after the input image is preprocessed; In step 2, the partitions where the salient objects with different salient ranks are located in the salient ranking map are weighted, the weight with the highest salient degree is 1, and along with the decrease of the salient degree ranking, the weight value calculation formula is as follows: , Wherein n is a ranking number, and beta is a test value; if a plurality of different significant objects with different significant ranks exist in a partition in a crossing way, the weight values of the backlight partition are accumulated and summed according to the weight values distributed by the pixel points where the different significant objects are located, and then the average value is calculated as the weight value of the partition, wherein the calculation formula is as follows: , Wherein: m×n is the resolution of the input image; r is the number of columns of the corresponding backlight partitions of the input image divided according to the number of the partitions of the backlight source, and S is the number of rows of the corresponding backlight partitions; the weight value for the row S and column R backlight partitions, For the weight value of the pixel points of the j-th row and i-th column of the image, M/R, N/S is the number of the pixel points in the backlight partition; In step 2, the final partition backlight value of each partition is calculated according to the following formula: , Wherein, the BL1 is the first partition backlight value; BL2 is the second partition backlight value; the weight values assigned to different partitions; BL4 is the final partition backlight value obtained, and all the partition backlight values obtained form an R×S backlight value matrix, wherein R represents the number of columns of the backlight partition, and S represents the number of rows of the backlight partition; And in the step 3, the dimming factors of the corresponding block backlight value modules are determined according to the final partition backlight values, pulse Width Modulation (PWM) signals of backlight LEDs in each block backlight value module are set according to the dimming factors and are sent to the LED driving modules of the corresponding backlight modules, a backlight matrix formed by all the final partition backlight values is subjected to fuzzy diffusion by adopting a BMA (body-building block) light mixing diffusion algorithm to obtain a fuzzy diffusion matrix, the backlight value corresponding to each pixel point on the liquid crystal panel is obtained by the fuzzy diffusion matrix, then the liquid crystal compensation signal of each pixel point is obtained by utilizing a liquid crystal compensation algorithm, and the liquid crystal compensation signals are sent to the liquid crystal control module, so that the display of an input image is carried out.
2. 2. The image saliency ranking model-based liquid crystal display area dynamic dimming method according to claim 1, wherein a liquid crystal compensation algorithm adopts a linear or nonlinear compensation algorithm.

Description

Liquid crystal display area dynamic dimming method based on image saliency ranking model Technical Field The invention relates to a liquid crystal display control method, in particular to a liquid crystal display area dynamic dimming method based on an image saliency ranking model. Background Liquid crystal display has become a mainstream flat panel display technology. In recent years, with the development of LCD display technology, a backlight module is continuously advancing, and common LED backlight approaches to a small-pitch LED and gradually develops towards a Mini-LED, and compared with the traditional direct type LED backlight, the Mini-LED backlight can realize more accurate regional light control, so that the screen brightness is more uniform, more partitions can be realized, and the display can realize better contrast, deeper black and lower power consumption. Dynamic dimming algorithms are also important for lower power consumption and higher display effect of the display. The dynamic dimming algorithm can dynamically change backlight brightness and liquid crystal pixel opening according to the display image, effectively improve display contrast and reduce power consumption. The existing dynamic dimming algorithm improves the quality of the display image to a certain extent, but has some defects. The average method has a good energy saving effect, but the display quality is reduced in the case of displaying the pixel values above the average value in the backlight brightness. The display effect of the image is not only related to the objective evaluation index, but also is closely related to the human visual sense effect. For an image, objective evaluation indexes such as PSNR, SSIM, information entropy and the like are good, and do not necessarily represent that the image display quality is better, and subjective evaluation of people is also important. There are many indices that combine subjective evaluation, and there are also people to begin research backlight algorithms based on human visual characteristics. The image quality objective evaluation index of the algorithm is probably inferior to that of other algorithms, but the display effect of the algorithm is good, so that the subjective feeling of a person is more comfortable. The dimming algorithm based on the image features can classify the images according to the detail features of the images, and adopts an optimal algorithm for different partitions, so that the image display quality is improved, and the energy consumption is saved. But this approach is only relevant to the human eye perception effect, irrespective of the importance of the image content and the attention mechanisms of the human brain. Therefore, a dynamic dimming algorithm capable of selecting an optimal dimming method according to the content importance of an image needs to be found, weight distribution is divided according to analysis of the image content, the fidelity of a key region image is improved, the subjective visual effect of a person is ensured, the display effect perceived by the human eye is improved to a greater extent, and the energy consumption is reduced. Liu et al, 2022, published in IEEE, instance-level saliency-ranking detection model, proposed graph-based reasoning, first split salient instances using an improved Mask R-CNN network, then add saliency-ranking branches to infer relative saliency. The relative saliency ranking builds a new graph inference module and also proposes a new penalty function to train the saliency ranking branches effectively. Finally, the authors built a new saliency ranking benchmark dataset from the MS-COCO dataset and SALICON dataset to train the saliency ranking model and validated the model being trained using the validation set. On the basis, the method is considered to be applied to a backlight algorithm image preprocessing part, the size of an input image is firstly adjusted, then the saliency scores of different salient objects in the input image are predicted to carry out salient ranking, and finally the size of an output image is restored to obtain a final saliency ranking map. The model can simulate the attention mechanism of the human brain, namely the staring time of the human eyes to different remarkable objects in the image, so as to distinguish the importance degree of each part and optimize the dimming method. Disclosure of Invention The invention provides a liquid crystal display area dynamic dimming method based on an image significance ranking model, which aims to solve the problem that the area dynamic dimming cannot be carried out from the attention level of human brain only from the human eye perception effect in the prior art. In order to achieve the above purpose, the technical scheme adopted by the invention is as follows: the liquid crystal display area dynamic dimming method based on the image saliency ranking model comprises the following steps: step 1, converting an input image into a gray le