Object Detection via Visual Search Similarity Voting Maps
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional sliding window approaches for object detection in images, such as face detection, struggle with variations like profile, slanted, or occluded views due to the difficulty in learning a good classifier encoding all possible variations.
Innovation Solution
Generating similarity voting maps by comparing a test image with multiple object images and using aggregate maps with gating and non-maxima suppression to improve detection accuracy and reduce false positives.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If sliding window approach is used for object detection, then detection speed is maintained, but detection accuracy deteriorates for challenging views (profile, slanted, occluded)
Solution Approach 1:
The patent pre-computes and stores multiple object images representing different views, orientations, and variations of the target object in a database before actual detection. During detection, the system retrieves relevant pre-computed object images and compares them with the test image using similarity voting maps, eliminating the need to learn all variations during training and enabling accurate detection of challenging views without sacrificing detection speed
2Measurement precision
If binary classifier is trained to encode all possible variations, then detection accuracy improves, but device complexity increases
Solution Approach 1:
Instead of training a complex binary classifier to encode all possible object variations, the patent creates a database of actual object images representing different views and variations. The detection system then uses these stored object images as references for similarity comparison through voting maps. This copying approach replaces the need for complex classifier training while maintaining high detection accuracy across various object presentations
3Measurement precision
If similarity voting maps with multiple object images are used, then detection accuracy improves, but processing time increases
Solution Approach 1:
The patent divides the detection process into independent segments: (1) pre-computing similarity maps between each object image and test image portions, (2) voting aggregation across multiple object images, and (3) threshold gating and non-maxima suppression. This segmentation allows parallel processing of multiple object images simultaneously and separates computation from decision-making, reducing overall processing time while maintaining high accuracy through aggregate similarity voting
Data Source
AI summary
One exemplary embodiment involves receiving a test image generating, by a plurality of maps for the test image based on a plurality of object images. Each of the object images comprises an object of a same object type, e.g., each comprising a different face. Each of the plurality of maps is generated to provide information about the similarity of at least a portion of a respective object image to each of a plurality of portions of the test image. The exemplary embodiment further comprises detecting a test image object within the test image based at least in part on the plurality of maps.