Two-Stage Image Segmentation for Clearer Foreground Boundaries
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing image segmentation methods suffer from low accuracy and information loss at boundaries between foreground and background regions, leading to suboptimal segmentation results.
Innovation Solution
A two-stage segmentation approach using a first segmentation model for preliminary recognition followed by a second model for optimization, combined with recombination of image regions to enhance boundary information, employing convolutional neural networks with depthwise separable convolutions to refine the segmentation process.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If a single segmentation model is used for image segmentation, then the processing speed is maintained, but the segmentation accuracy and boundary clarity deteriorate
Solution Approach 1:
The patent divides the segmentation process into two independent stages: a first segmentation model performs initial segmentation to obtain candidate foreground and background regions, while a second segmentation model performs refined segmentation on the recombined image. This multi-stage segmentation approach improves boundary accuracy without requiring a single overly complex model, as each stage has specialized functionality.
Solution Approach 2:
The first segmentation model performs preliminary segmentation recognition to obtain candidate regions before the second model processes the image. This preliminary action provides initial segmentation results that guide subsequent refined processing, allowing the system to achieve high accuracy through progressive refinement rather than requiring all complexity in one model.
2Reliability
If traditional segmentation methods are used, then the processing is simple, but information loss at boundaries occurs leading to poor segmentation results
Solution Approach 1:
The patent recombines the candidate foreground image region and candidate background image region to form a recombined image that preserves boundary information from both segments. This merging operation ensures that boundary regions are not lost during segmentation, as they appear in both foreground and background candidates and are thus retained in the recombined output for further processing.
Solution Approach 2:
The system uses the candidate regions from the first segmentation model as feedback to guide the second segmentation model. The second model processes the recombined image with awareness of the initial segmentation results, allowing iterative refinement that prevents information loss by continuously referencing previous segmentation outcomes.
Data Source
Figure 1
Figure 2a
Figure 2b
AI summary
Disclosed are an image processing method and apparatus, and a device and a storage medium, which belong to the technical field related to computer vision in artificial intelligence. The method comprises: performing preliminary segmentation recognition on an original image by using a first segmentation model, so as to obtain a candidate foreground image area and a candidate background image area of the original image; recombining the candidate foreground image area, the candidate background image area and the original image to obtain a recombined image, wherein pixel points in the recombined image correspond to pixel points in the original image on a one-to-one basis; and performing area segmentation recognition on the recombined image by using a second segmentation model, so as to obtain a target foreground image area and a target background image area of the original image.