Face Detection Using Rotated Sample Images and Universal Detectors
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing face detection methods require extensive preparation of sample image groups for various orientations and directions, leading to inefficient learning and detection processes, especially when dealing with multiple combinations of face directions and orientations.
Innovation Solution
A method that generates face sample image groups by applying mirror reversal processing and 90-degree rotation to initial groups, allowing detectors to learn characteristics for a broader range of orientations and directions, thereby reducing the need for multiple detectors and shortening the learning time.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If multiple detectors are generated for each combination of face directions and orientations, then detection accuracy for various face orientations is improved, but preparation time and learning efficiency deteriorate
Solution Approach 1:
A single detector is designed to perform multiple functions by detecting faces in various directions and orientations simultaneously. The detector uses multiple detection regions (first detection region for front faces, second detection region for profile faces, third detection region for three-quarter faces) within one unified structure, eliminating the need for separate detectors for each face orientation while maintaining comprehensive detection accuracy.
Solution Approach 2:
The detection process is segmented into different detection regions within a single detector. The first detection region handles front faces, the second detection region handles profile faces, and the third detection region handles three-quarter faces. This segmentation allows one detector to efficiently handle multiple face orientations without requiring separate detectors for each type.
2Adaptability or versatility
If multiple detectors are generated for each combination of face directions and orientations, then detection coverage is improved, but device complexity increases
Solution Approach 1:
One universal detector is designed to detect faces in all directions and orientations by incorporating multiple detection regions. The first detection region detects front faces, the second detection region detects profile faces, and the third detection region detects three-quarter faces. This multi-functional design reduces the number of detectors from multiple specialized detectors to a single versatile detector, thereby reducing device complexity while maintaining comprehensive detection coverage.
Solution Approach 2:
Multiple detection functions that would traditionally require separate detectors are merged into a single detector. The detector combines the first detection region for front faces, second detection region for profile faces, and third detection region for three-quarter faces into one unified detection apparatus, simplifying the overall system structure while preserving the ability to detect various face orientations.
3Measurement precision
If learning is carried out for each combination of face directions and orientations, then detection accuracy is improved, but learning efficiency deteriorates
Solution Approach 1:
The learning process is optimized by training a single universal detector to recognize faces in various directions and orientations simultaneously. The detector learns from training images that include different face orientations (front, profile, three-quarter faces) and develops the capability to detect all these orientations within one learning cycle, rather than requiring separate learning processes for each orientation combination. This significantly improves learning efficiency while maintaining high detection accuracy across all face orientations.
Data Source
AI summary
In a method of detecting a face in various directions in a target image with use of detectors, a partial image cut sequentially from the target image is subjected to mirror reversal processing and rotation processing in 90 degree increments for generating reversed/rotated images of the partial image. The detectors of predetermined types judge whether the respective images represent face images in predetermined face directions and orientations. Based on combinations of the types of the detectors and the types of the input images, faces in various face directions and orientations can be judged.