Search

CN-122021761-A - Training system, training method, dimming system and dimming method

CN122021761ACN 122021761 ACN122021761 ACN 122021761ACN-122021761-A

Abstract

A training system, training method, dimming system, dimming method, computer readable recording medium having stored program and non-transitory computer program product, wherein the training method is used for training a neural network module to be trained, executed by a processing module, the neural network module to be trained comprises a target image generation module, a neural network to be trained and a light distribution generation module, the training method is executed in one training round, the training method comprises repeatedly executing training images in a training set as input images, performing convolution operation on intermediate compensation images and light distribution of the training images to generate convolution images, and obtaining losses based on the convolution images and the target images, and updating a plurality of parameters based on an average and updating algorithm of all the losses obtained in the foregoing steps.

Inventors

  • WANG ZHENGJUN

Assignees

  • 瑞昱半导体股份有限公司

Dates

Publication Date
20260512
Application Date
20241111

Claims (10)

  1. 1. A training system comprising a processing module and a neural network module to be trained, wherein the neural network module to be trained comprises: A target image generation module configured to receive an input image and obtain a target image based on a target calculation program; a neural network to be trained having a plurality of parameters, the neural network to be trained configured to generate an intermediate compensated image based on the input image and the plurality of parameters, and A light distribution generating module configured to generate a light distribution based on the input image, and The processing module is configured to perform in a training round: (a) Repeatedly performing a convolution operation on the intermediate compensated image and the light distribution of the training image to generate a convolution image, taking a training image in a training set as the input image, and obtaining a loss based on the convolution image and the target image, and (B) Updating the plurality of parameters based on an average of all of the losses obtained in step (a) and an updating algorithm.
  2. 2. The training system of claim 1, wherein the target image generation module comprises a depth message model, the target calculation program comprising (c) obtaining a depth image corresponding to the input image based on the depth message model, and (d) adjusting the input image to obtain the target image based on depth information of the depth image.
  3. 3. The training system of claim 2, wherein said step (d) Comprising: (d1) Based on the brightness adjusting coefficient, the brightness of the input image is adjusted to be high so as to obtain a brightness-enhanced image; (d2) Adjusting the gamma value of the input image via an S-curve to obtain an enhanced contrast image, and (D3) The method includes obtaining a depth-adjusted enhanced contrast image by performing a point-wise multiplication operation on the depth image and the enhanced contrast image, obtaining a difference tensor by subtracting the depth image from a full tensor, obtaining a depth-adjusted enhanced image by performing the point-wise multiplication operation on the difference tensor and the enhanced image, and obtaining the target image by performing a point-wise addition operation on the depth-adjusted enhanced contrast image and the depth-adjusted enhanced image.
  4. 4. The training system of claim 2, wherein the depth message model comprises a MiDaS model.
  5. 5. The training system of claim 1, wherein the loss employs a mean square error.
  6. 6. The training system of claim 1, wherein the light distribution generation module comprises a backlight decision module configured to receive the input image and to generate a plurality of backlight intensities based on the input image, the light distribution generation module configured to generate the light distribution based on the plurality of backlight intensities.
  7. 7. A training method is used for training a neural network module to be trained, and is executed by a processing module, wherein the neural network module to be trained comprises a target image generating module, a target image generating module and a training module, wherein the target image generating module is configured to receive an input image and obtain a target image based on a target computing program, the neural network to be trained comprises a plurality of parameters, The neural network to be trained is configured to generate an intermediate compensated image based on the input image and the plurality of parameters, and a light distribution generation module configured to generate a light distribution based on the input image, the training method including performing in a training round: (a) Repeatedly performing a convolution operation on the intermediate compensated image and the light distribution of the training image to generate a convolution image, taking a training image in a training set as the input image, and obtaining a loss based on the convolution image and the target image, and (B) Updating the plurality of parameters based on an average of all of the losses obtained in step (a) and an updating algorithm.
  8. 8. A dimming system trained using the training system of claims 1-6 to obtain the plurality of parameters, comprising: A dimming system backlight decision module configured to receive an image and generate a plurality of backlight intensities of the image based on the image, and A neural network module including a neural network configured to have the same architecture as the neural network to be trained, the neural network module configured to store the plurality of parameters and configured to receive the image and generate a compensated image of the image based on the neural network and the plurality of parameters.
  9. 9. The dimming system of claim 8, wherein the dimming system comprises: a backlight driving module configured to receive the plurality of backlight intensities and drive a backlight module of a display based on the plurality of backlight intensities, and The panel driving module is configured to receive the compensation image and drive a display panel of the display based on the compensation image.
  10. 10. A dimming method for training the plurality of parameters obtained using the training method of claim 7, comprising: Receiving an image by a dimming system backlight decision module and generating a plurality of backlight intensities of the image based on the image, and The image is received by a neural network module, and a compensated image of the image is generated based on the neural network and the plurality of parameters, wherein the neural network has the same architecture as the neural network to be trained.

Description

Training system, training method, dimming system and dimming method Technical Field The present invention relates to the field of dimming. In particular to a technology for applying a neural network to dimming. Background In the current Local dimming (Local dimming) system, the process is generally first to make a backlight decision (Backlight decision), and the backlight intensity of each block is determined according to the algorithm design, and sometimes the maximum pixel is used and sometimes the average pixel is used. Light spread Modeling (LIGHT SPREAD Modeling) is followed, primarily to calculate the light distribution of the backlight in terms of backlight intensity (Light Distribution). Pixel compensation (Pixel Compensation) is then performed to adjust each pixel according to the backlight intensity to maintain image stability. Ideally, this increases the image contrast. There are problems associated with the above-described adjustment of the backlight and the corresponding pixel compensation, and first, in some low-brightness scenes, black side effects may occur in the conventional manner because the pixel compensation is calculated by using the regional backlight to compensate. In addition, the pixel compensation is determined according to the area backlight, and the process lacks a depth of field concept, so that the farther the image picture is, the darker the depth of field effect is. Disclosure of Invention In view of the above, some embodiments of the present invention provide a training system, training method, dimming system, dimming method, computer readable recording medium having stored program, and non-transitory computer program product to improve the prior art. Some embodiments of the present invention provide a training system comprising a processing module and a neural network module to be trained, wherein the neural network module to be trained comprises a target image generation module configured to receive an input image and to obtain the target image based on a target calculation program, a neural network to be trained having a plurality of parameters, the neural network to be trained configured to generate an intermediate compensated image based on the input image and the plurality of parameters, and a light distribution generation module configured to generate a light distribution based on the input image, and the processing unit is configured to perform in one training round repeatedly performing a training image in the training set as the input image, performing a convolution operation on the intermediate compensated image and the light distribution of the training image to generate a convolution image, and obtaining a loss based on the convolution image and the target image, and updating the plurality of parameters based on an average and updating algorithm of all losses obtained in the foregoing steps. Some embodiments of the present invention provide a training method for training a neural network module to be trained, performed by a processing module, the neural network module to be trained comprising a target image generation module configured to receive an input image and to obtain a target image based on a target calculation program, the neural network to be trained having a plurality of parameters, the neural network to be trained configured to generate an intermediate compensated image based on the input image and the plurality of parameters, and a light distribution generation module configured to generate a light distribution based on the input image, the training method comprising performing in one training round repeatedly performing a training image in the training set as the input image, performing a convolution operation on the intermediate compensated image and the light distribution of the training image to generate a convolution image, and obtaining a loss based on the convolution image and the target image, and updating the plurality of parameters based on an average and updating algorithm of all losses obtained in the preceding steps. Some embodiments of the present invention provide a dimming system comprising a dimming system backlight decision module configured to receive an image and generate a plurality of backlight intensities of the image based on the image, and a neural network module comprising a neural network configured to have the same architecture as the neural network to be trained, the neural network module configured to store parameters obtained by training of the training system and configured to receive the image, and generate a compensated image of the image based on the neural network and the plurality of parameters. Some embodiments of the present invention provide a dimming method, which includes receiving an image by a backlight decision module of a dimming system and generating a plurality of backlight intensities of the image based on the image, and receiving the image by a neural network module and generating a compensation im