Face Detection Using Patch-Based Bounding Box Estimation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current face detection algorithms are computationally expensive and require high-end hardware, leading to increased processing time and CPU usage, even with high-definition images, while existing techniques struggle to maintain accuracy at higher pixel shifts and grid spacings.
Innovation Solution
A face-detection processing method that uses a patch-based approach with a decision tree classifier to estimate bounding boxes at higher pixel shifts, reducing the number of windows scanned and employing a 6×6 grid spacing for initial patch classification, followed by localized search within the bounding box using an aggressive 1×1 grid spacing.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If a sliding window approach with fine grid spacing is used for face detection, then detection accuracy is improved, but processing time and computational cost increase significantly
Solution Approach 1:
The patent divides the face detection task into two stages: first using a patch-based approach with 6×6 grid spacing to identify candidate regions, then performing localized search with 1×1 grid spacing only within those candidate regions. This segmentation of the search space dramatically reduces the total number of windows scanned while maintaining detection accuracy.
Solution Approach 2:
The patent performs preliminary classification using a decision tree classifier on patches extracted at 6×6 grid spacing before conducting the full face detection. This preliminary action identifies candidate bounding boxes that contain faces, allowing the subsequent fine-grid search to be confined to these limited regions rather than the entire image.
2Productivity
If high-end hardware is used for face detection processing, then processing speed is improved, but device cost and complexity increase
Solution Approach 1:
The patent changes the grid spacing parameter dynamically based on the detection stage: using 6×6 grid spacing for initial patch classification and decision tree evaluation, then switching to 1×1 grid spacing for localized face detection within candidate regions. This parameter optimization enables fast processing on low-cost hardware.
Solution Approach 2:
The patent extracts only the essential features needed for initial classification using a decision tree on patches, rather than applying computationally expensive classifiers to every possible window position. This extraction of key information reduces computational burden and enables deployment on hardware with limited processing capability.
3Productivity
If grid spacing is increased to reduce the number of windows scanned, then processing speed is improved, but detection accuracy decreases
Solution Approach 1:
The patent employs dynamic grid spacing that adapts to the detection context: using coarser 6×6 grid spacing when evaluating patches for candidate region identification, then switching to finer 1×1 grid spacing within identified candidate bounding boxes. This dynamic adjustment maintains both speed and accuracy.
Solution Approach 2:
The patent applies different grid spacing qualities to different regions of the image: coarse 6×6 grid spacing is applied to the entire image for patch evaluation, while fine 1×1 grid spacing is applied only within candidate bounding boxes identified as containing faces. This local quality differentiation optimizes both processing efficiency and detection accuracy.
Data Source
AI summary
A method for detecting a presence of at least one face in at least one image is comprises creating an image patch map based on a plurality of face patches identified for at least one window in the in at least one image, estimating a bounding box, and searching within the bounding box to detect presence of the at least one face in the at least one image. The present disclosure discloses use of any classifier which works on top of any feature representation to identify face patches and then using a masking system to identify bounding boxes.