Silhouette Image Processing Using Neural Networks
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing silhouette operations in electronic devices require high hardware configuration, consume significant power due to extensive graphics operations, and are limited by system platforms, particularly with the ARKit platform being closed-source and only applicable to iOS 11 or later, excluding Android and devices without depth-of-field cameras.
Innovation Solution
A method using a trained neural network model to recognize and separate image regions of interest without depth-of-field information, applicable across various operating systems, including Android, and reducing energy consumption by converting image form to array form for processing.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If depth-of-field camera and ARKit platform are used for silhouette operations, then silhouette extraction accuracy is improved, but device compatibility is reduced and hardware requirements are increased
Solution Approach 1:
The patent replaces the depth-of-field camera hardware system with a neural network-based image processing system. Instead of relying on optical depth information from specialized cameras, the invention uses a trained neural network model to analyze standard images and extract silhouette information, thereby eliminating hardware dependencies and expanding device compatibility while maintaining extraction accuracy
Solution Approach 2:
The patent changes the input parameters from depth-map images (requiring specialized hardware) to standard RGB images (universally captured). By training the neural network to process conventional image data instead of depth information, the system achieves the same silhouette extraction goal using parameters available on all modern devices, thus improving compatibility without sacrificing accuracy
2Manufacturing precision
If extensive graphics operations are performed for silhouette operations, then silhouette extraction quality is improved, but energy consumption is increased
Solution Approach 1:
The patent performs silhouette extraction in advance during the image processing pipeline, before the extracted regions need to be applied. By pre-processing images to identify and separate silhouette regions using the neural network, the system avoids performing computationally intensive graphics operations later when the results are actually needed, thereby reducing overall energy consumption while maintaining extraction quality
Solution Approach 2:
The patent extracts only the essential silhouette region information from images using the neural network, rather than performing extensive graphics operations on entire images. This selective extraction approach focuses computational resources on identifying and isolating relevant regions, significantly reducing the total energy required for silhouette operations while preserving extraction quality
3Ease of manufacture
If ARKit platform is used for silhouette operations, then silhouette extraction functionality is achieved, but system platform dependency is increased
Solution Approach 1:
The patent develops a universal neural network model that can run on multiple operating systems and device types without requiring platform-specific frameworks like ARKit. The system processes images and extracts silhouettes using the same core algorithm across different platforms, making the functionality portable and eliminating system dependency while maintaining ease of implementation
Solution Approach 2:
The patent creates a software-based neural network solution that replicates the silhouette extraction functionality previously available only through proprietary platform frameworks. By copying and adapting the core extraction capability into a platform-agnostic neural network, the invention achieves the same functional results without requiring specialized system support, thereby reducing device complexity and platform dependency
Data Source
AI summary
A method of image processing includes: obtaining an original image and obtaining a to-be-processed image based on the original image; recognizing the to-be-processed image by using an image recognition model to obtain a silhouette array; generating a silhouette image based on the silhouette array; and generating a target image based on the original image and the silhouette image, where the target image corresponds to a region of interest of the original image.