Image Segmentation via Mask Quality Unification
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing edge-based image segmentation methods are inefficient due to high user participation requirements, unsatisfactory results, and high costs associated with training deep-learning models using finely labeled sample images.
Innovation Solution
An image processing method involving a mask prediction model to obtain initial contour data, a mask quality unification model to adjust contour fineness, and a target image obtaining model to refine the segmentation, reducing the need for high-fineness contour data and extensive training data.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If deep-learning-based image matting is used to achieve fine segmentation, then segmentation precision is improved, but training costs increase due to requiring finely labeled sample images
Solution Approach 1:
The patent uses coarse contour data instead of expensive fine contour data for training the mask prediction model. This substitutes high-cost training data with low-cost alternatives, achieving satisfactory segmentation results without requiring finely labeled sample images, thereby reducing training complexity and costs.
Solution Approach 2:
The patent introduces a mask quality unification model as an intermediary between the mask prediction model and the final segmentation result. This intermediary model refines the coarse mask output by adjusting mask quality based on contour fineness, enabling the system to achieve fine segmentation results without requiring fine contour training data, thus resolving the contradiction between precision and training complexity.
2Measurement precision
If interactive image matting is used to obtain fine segmentation, then segmentation precision is improved, but processing time increases due to requiring large number of users
Solution Approach 1:
The patent employs automated models (mask prediction model and mask quality unification model) that process images independently without requiring user interaction. The system automatically predicts masks from coarse contour data and refines them through the unification model, achieving fine segmentation results without manual intervention, thereby eliminating the time loss associated with interactive methods.
Solution Approach 2:
The patent performs preliminary downsampling of the input image to obtain coarse contour data before processing. This preliminary action reduces the complexity of subsequent processing steps, enabling the automated models to work efficiently on simplified data while still producing fine segmentation results, thus reducing overall processing time compared to interactive methods.
3Productivity
If smart image matting is used to reduce user participation, then processing efficiency is improved, but segmentation precision deteriorates
Solution Approach 1:
The mask quality unification model serves as an intermediary that enhances the output of the automated mask prediction model. It refines the coarse mask by adjusting quality parameters based on contour fineness, thereby improving segmentation precision while maintaining the high processing efficiency of automated methods, resolving the contradiction between efficiency and precision.
Solution Approach 2:
The patent changes the parameter of contour fineness by processing the image at different resolutions. It downsamples the image to obtain coarse contour data for efficient processing, then uses the unification model to adjust mask quality parameters based on the actual contour fineness required, achieving both high efficiency and improved precision.
Data Source
AI summary
Image processing methods, devices, and storage media are provided. One of the image processing methods includes: obtaining an image comprising a target object; inputting the image into a mask prediction model to obtain a first mask image corresponding to the target object in the image; inputting the first mask image and the image comprising the target object into a mask quality unification model to obtain a second mask image, wherein the mask quality unification model is configured to adjust a fineness associated with the first mask image to a target fineness to obtain the second mask image, and wherein the second mask image has image semantic information consistent with image semantic information of the image comprising the target object; and inputting the image comprising the target object and the second mask image into a target image obtaining model to obtain a target image corresponding to the target object.