Face Search Accuracy With Inhibiting Factor Detection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional people search systems face reduced search accuracy due to inhibiting factors like masks or sunglasses, leading to increased search errors when similarity thresholds are lowered to compensate, which decreases efficient search performance.
Innovation Solution
A people search system that detects inhibiting factors in captured face images and generates composite face images by combining these factors with registered face images, allowing for feature extraction and similarity-based searches, thereby maintaining search accuracy and efficiency.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If the similarity threshold is reduced to prevent lower search accuracy caused by inhibiting factors, then search accuracy is improved, but search errors increase
Solution Approach 1:
The system performs preliminary detection of inhibiting factors (masks, sunglasses, etc.) in the surveillance image before conducting the similarity search. Based on this detection, it pre-processes the registered face image by adding the same inhibiting factors to generate a modified registered image that matches the surveillance conditions, thereby eliminating the need to lower the similarity threshold and avoiding search errors.
Solution Approach 2:
The system converts the harmful effect of inhibiting factors (which normally reduce search accuracy) into a benefit by using these same factors to generate a modified registered image. The inhibiting factors detected in the surveillance image are applied to the registered face image to create a version that accounts for the masking effects, thereby improving search accuracy without increasing error rates.
2Measurement precision
If the similarity threshold is reduced to account for inhibiting factors, then search accuracy is improved, but search efficiency decreases
Solution Approach 1:
The system performs preliminary detection of inhibiting factors and pre-processing of the registered image before the similarity search. This preliminary action ensures that the registered image already accounts for masking effects, allowing the use of a normal similarity threshold and maintaining high search efficiency without sacrificing accuracy.
Solution Approach 2:
The system introduces an intermediary processing step that generates a modified registered image by adding detected inhibiting factors to the original registered face image. This intermediary image serves as a bridge between the original registered image and the surveillance image with masking, enabling accurate comparison at normal similarity thresholds and maintaining search efficiency.
3Productivity
If the registered face image is used without modification, then search efficiency is maintained, but search accuracy decreases due to inhibiting factors
Solution Approach 1:
The system performs preliminary detection of inhibiting factors in the surveillance image and applies the same factors to the registered face image before the similarity search. This preliminary modification ensures that the registered image reflects the actual viewing conditions, improving search accuracy while maintaining efficiency through a single-pass search at normal thresholds.
Solution Approach 2:
The system converts the harmful masking effects present in the surveillance image into a benefit by applying the same masks to the registered face image. This creates a modified registered image that accounts for the inhibiting factors, thereby improving search accuracy without compromising search efficiency.
Data Source
AI summary
According to an embodiment, a people search system includes a receiver, a detector, a search unit, and a display. The search unit obtains features of a first face image and a second face image of a predetermined person and searches for the predetermined person based on a similarity between the features of the first and the second face images when an inhibiting factor is not detected, and generates a third face image that is a composite image of the second face image and the inhibiting factor, obtains features of the first and the third face images, and searches for the predetermined person based on a similarity between the features of the first and the third face images when the inhibiting factor is detected. The display displays the search result of the predetermined person obtained by the search unit.