DNN Image Matting via Subnetwork Trimap Generation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing photo manipulation tools are tedious to use, require significant skill, and often produce inaccurate results when selecting objects from images with complex backgrounds and fine details like hair or fur, making image matting a challenging task.
Innovation Solution
A deep neural network (DNN)-based system that generates an accurate alpha matte from user-provided selection indicators, utilizing a three-subnetwork approach to create a binary mask, trimap, and alpha matte, with techniques for improved computational efficiency and decontamination of foreground and background colors to minimize halo effects.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If existing photo manipulation tools are used for image matting, then object selection can be performed, but the process becomes tedious and requires significant skill while producing inaccurate results
Solution Approach 1:
The system performs automatic image matting by analyzing the input image and autonomously generating the alpha matte without requiring manual user intervention. The neural network automatically identifies object boundaries, handles complex backgrounds, and produces accurate matting results, making the tool self-sufficient and eliminating the need for skilled manual operation.
Solution Approach 2:
The patent replaces manual photo manipulation tools with an automated neural network system. Instead of requiring users to manually select and mask objects through complex tool operations, the system uses deep learning algorithms to automatically perform image matting, substituting mechanical user actions with intelligent automated processing.
2Manufacturing precision
If existing tools are used to select objects with fine details like hair or fur against complex backgrounds, then object extraction is attempted, but inaccurate results are produced
Solution Approach 1:
The system changes the operational parameters from manual tool-based selection to neural network-based processing. The neural network uses learned features and probabilistic modeling to accurately determine pixel-level alpha values, particularly for challenging regions with fine details like hair or fur against complex backgrounds, achieving both high boundary accuracy and reliable results.
3Measurement precision
If a three-subnetwork approach is used to generate binary mask, trimap, and alpha matte, then matting accuracy is improved, but computational complexity increases
Solution Approach 1:
The patent segments the image matting task into three distinct subnetworks: a binary mask generation network, a trimap generation network, and an alpha matte generation network. Each subnetwork focuses on a specific aspect of the matting process, allowing for specialized optimization and improved overall accuracy while maintaining modular architecture that manages computational complexity.
Solution Approach 2:
The trimap serves as an intermediary output from the second subnetwork that guides the third subnetwork in generating the final alpha matte. This intermediate representation captures uncertain regions and provides structured information that improves the accuracy of the final matting result while organizing the computational flow across multiple networks.
Data Source
AI summary
Techniques are disclosed for deep neural network (DNN) based interactive image matting. A methodology implementing the techniques according to an embodiment includes generating, by the DNN, an alpha matte associated with an image, based on user-specified foreground region locations in the image. The method further includes applying a first DNN subnetwork to the image, the first subnetwork trained to generate a binary mask based on the user input, the binary mask designating pixels of the image as background or foreground. The method further includes applying a second DNN subnetwork to the generated binary mask, the second subnetwork trained to generate a trimap based on the user input, the trimap designating pixels of the image as background, foreground, or uncertain status. The method further includes applying a third DNN subnetwork to the generated trimap, the third subnetwork trained to generate the alpha matte based on the user input.