Image processing model training method, image processing method and device
By processing color-corrected image data and mask images using a neural network model, high-frequency and low-frequency component information are separated, solving the problem of poor image quality in mobile terminal flash scenarios and improving image quality and aesthetics.
CN122157331APending Publication Date: 2026-06-05VIVO MOBILE COMM CO LTD
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- VIVO MOBILE COMM CO LTD
- Filing Date
- 2026-03-10
- Publication Date
- 2026-06-05
AI Technical Summary
Technical Problem
The flash module of the mobile terminal cannot dynamically adjust according to the shooting distance, resulting in poor image quality when shooting in flash scenes, such as overexposed and distorted faces or insufficient lighting and loss of details.
Method used
By acquiring a training sample set, including color-corrected image data and mask images, a neural network model is trained to separate and process the high-frequency and low-frequency component information of the image, and the detail weight image is fused to improve image quality.
Benefits of technology
It significantly improves the aesthetics of photos taken under flash, preserving details of people and natural lighting effects, while reducing the need for flash power.
✦ Generated by Eureka AI based on patent content.
Smart Images
Figure CN122157331A_ABST
Abstract
The application discloses an image processing model training method, an image processing method and device, and belongs to the technical field of image processing. The method comprises the following steps: acquiring a first training sample set, wherein the first training sample set comprises multiple groups of first training samples, one group of first training samples comprises first sample image data, a sample mask image and a first reference image corresponding to the first sample image data, the sample mask image is determined based on a portrait semantic information region in the first sample image data, the first sample image data is obtained by performing down-sampling on image data obtained by performing color correction on second sample image data, the second sample image data is RAW format image data collected by an electronic device in an opened state of a flash, and the first reference image is obtained by performing down-sampling on a second reference image, and the second reference image is an image collected by an image collection device; and training a first neural network model by using the first training sample set to obtain a second neural network model.
Need to check novelty before this filing date? Find Prior Art