Method, device and equipment for region segmentation and hair replacement of video images
By deploying a lightweight deep convolutional neural network on the user terminal, using the encoder and decoder modules to extract image features and generate probabilistic images, the existing technology problems of hair segmentation models occupying large memory and low segmentation accuracy are solved. Real-time and accurate hair area segmentation.
CN112200816BActive Publication Date: 2025-01-24GUANGZHOU HUADUO NETWORK TECH
View PDF 0 Cites 0 Cited by
Patent Information
- Application Number
- CN202011104747.3
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2020-10-15
- Publication Date
- 2025-01-24
- Estimated Expiration
- 2040-10-15
AI Technical Summary
Technical Problem
In the prior art, the hair segmentation model occupies a large amount of memory, which makes it impossible to deploy on the user terminal to run in real time, and the segmentation accuracy is low.
Method used
A lightweight deep convolutional neural network is used as the hair segmentation network. Image features of different receptive fields are extracted through the encoder and decoder modules, and feature jumping and upsampling techniques are used to restore the feature images and generate probabilistic images for hair region segmentation.
Benefits of technology
It achieves real-time and accurate hair area segmentation of video images on the user terminal, reduces the amount of calculation and memory usage, and improves the segmentation accuracy.
✦ Generated by Eureka AI based on patent content.
Smart Images
Figure CN112200816B_ABST
Abstract
The method, device, equipment and computer-readable storage medium for regional segmentation and hair replacement of video images provided in the embodiments of the present application relate to the field of image processing. The method obtains a set of images to be segmented corresponding to an image frame sequence of a video image; uses an encoder module of a hair segmentation network to extract first image features with different receptive fields of the image to be segmented, inputs the first image features to a decoder module of a hair segmentation network for processing, restores a feature image consistent with the resolution of the input image to be segmented, obtains a probability image corresponding to the image to be segmented, and segments the hair region of the image frame sequence, wherein the hair segmentation network is a lightweight deep convolutional neural network deployed on a client. The technical solution realizes real-time and accurate segmentation of the hair region of a video image and real-time hair style transformation on a client, thereby improving the display effect of the video image.
Need to check novelty before this filing date? Find Prior Art