Face Detection Probability Index Filtering for Accuracy and Speed
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing face detection methods struggle with accurately identifying face images in low-quality images with uneven brightness or contrast, and are prone to false positives due to similar non-face images, while also being computationally inefficient when considering multiple positions and inclinations.
Innovation Solution
A method that cuts out fraction images at various positions and inclinations, calculates a probability index, extracts images with indices above a threshold, and increases the index for images similar in inclination, detecting faces efficiently by limiting subsequent calculations to confirmed inclinations, thereby reducing false detections and improving processing speed.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If face detection is performed by calculating probability indices for multiple positions and inclinations, then detection accuracy is improved, but processing time increases significantly
Solution Approach 1:
The patent performs preliminary face detection without inclination consideration first, then only performs inclination-specific detection for regions that initially show face-like characteristics. This preliminary filtering reduces the number of regions requiring exhaustive inclination analysis, thereby reducing processing time while maintaining detection accuracy for actual faces.
Solution Approach 2:
The detection process is segmented into two stages: a first detection stage that identifies candidate regions with face-like probability indices, and a second detection stage that performs detailed inclination analysis only on these candidate regions. This segmentation avoids unnecessary calculations for non-face regions, resolving the contradiction between comprehensive detection and processing efficiency.
2Productivity
If face detection is performed by comparing probability indices with threshold values, then simple processing is achieved, but false detections occur in low-quality images with uneven brightness or contrast
Solution Approach 1:
The patent uses detection results from the first detection stage as feedback to guide the second detection stage. Regions identified as having face-like characteristics trigger more rigorous inclination-based verification, while non-face regions are quickly dismissed. This feedback mechanism improves reliability for low-quality images without significantly impacting overall processing speed.
3Measurement precision
If exhaustive inclination analysis is performed for all detected faces, then accurate face orientation identification is achieved, but processing becomes computationally expensive
Solution Approach 1:
The patent applies partial action by performing exhaustive inclination analysis only on candidate regions that pass the initial face-like probability threshold, rather than on all detected regions. This selective application of detailed analysis maintains orientation accuracy for potential faces while avoiding computational waste on non-face regions, thus improving processing throughput.
Data Source
AI summary
An index representing the probability that a fraction image is a face image including a face in an input image is calculated for each of the positions of the face to be detected on the basis of a feature value. When the sum of the indexes of the fraction images is not smaller than the first threshold value, the image formed by the fraction images is determined to be a face image.