Image Parsing Method Using Segmented Feature Channels
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current facial recognition technologies face challenges due to the large volume and space occupation of image parsing models, which are caused by using multiple channels in different fields of view for feature extraction and parsing, leading to inefficiency and increased complexity.
Innovation Solution
An image parsing method that divides feature information into two categories based on preset size ranges and uses corresponding channels to parse each category, reducing the number of parsing channels and complexity, and adaptively adjusts the field of view sizes during training to improve accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If multiple channels in different fields of view are used for feature extraction and parsing, then parsing accuracy is improved, but model volume and space occupation increase excessively
Solution Approach 1:
The patent segments the feature information into two distinct groups: first feature information with smaller size ranges and second feature information with larger size ranges. Each group is processed by dedicated parsing channels, replacing the traditional multi-channel approach with a segmented two-channel architecture that reduces model volume while maintaining parsing accuracy
Solution Approach 2:
The patent applies local quality by assigning different processing characteristics to different feature size ranges. Small-sized features (first feature information) and large-sized features (second feature information) are handled by channels optimized for their respective size characteristics, improving parsing accuracy without requiring all channels to process all feature types
2Reliability
If multiple channels in different fields of view are used for feature parsing, then parsing completeness is improved, but device complexity increases
Solution Approach 1:
The patent divides the complex multi-channel parsing process into two segmented parsing channels, each handling specific feature size ranges. This segmentation simplifies the overall device complexity by reducing the number of channels from multiple to two, while maintaining parsing completeness through dedicated processing for each feature category
Solution Approach 2:
The patent changes the parameter of field of view size by adaptively adjusting it during training based on feature information size. This parameter adjustment allows the system to maintain high parsing completeness without requiring fixed complex multi-channel architectures, thereby reducing device complexity
3Loss of information
If feature information is processed through multiple parsing channels, then feature coverage is improved, but processing efficiency decreases
Solution Approach 1:
The patent segments feature information processing into two efficient pathways based on size ranges, allowing parallel processing of first and second feature information. This segmentation maintains comprehensive feature coverage while improving parsing efficiency by avoiding the bottlenecks of processing all features through multiple sequential channels
Solution Approach 2:
The patent extracts and separates first feature information and second feature information into distinct processing streams, taking out the inefficiency of unified multi-channel processing. This extraction enables targeted processing that improves overall parsing efficiency while maintaining complete feature coverage
Data Source
AI summary
An image parsing method includes obtaining feature information of an initial image, parsing first feature information in the feature information using a first channel to obtain a first prediction result, parsing second feature information in the feature information using a second channel to obtain a second prediction result, where a size of the first feature information meets a first size range, a size of the second feature information meets a second size range, and the first size range is less than the second size range, and outputting the first prediction result and the second prediction result as a parsing result of the initial image.