Orientation Independent Face Detection via Multi-Oriented Edge Detectors
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current face detection methods face challenges with orientation independence, requiring high computation and suffering from limited performance, especially for multi-view face detection and in-plane rotation, leading to increased false alarms.
Innovation Solution
A method employing multiple mosaic images with differently oriented edge detectors for edge detection, followed by a structure-based detection scheme using Adaboost for verification, allowing face detection over a 360-degree span of in-plane rotation angles, reducing false alarms through multi-scale and multi-level processing.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If statistical methods are used for face detection by scanning images on multiple scales, then detection capability is improved, but computation capability requirement increases
Solution Approach 1:
The patent segments the face detection process into two distinct stages: a structure-based detection stage that identifies candidate regions using geometric constraints, and a statistical verification stage that confirms faces using Adaboost classifiers. This segmentation allows the computationally intensive statistical methods to be applied only to selected candidate regions rather than the entire image, thereby reducing overall computation requirements while maintaining detection capability.
Solution Approach 2:
The patent performs preliminary structure-based detection to identify candidate face regions before applying statistical verification methods. By pre-processing the image to locate potential face candidates using efficient structure analysis, the system prepares the data in advance so that subsequent statistical verification operates on a reduced set of candidates, reducing computation burden while preserving detection reliability.
2Productivity
If structure-based methods are used for face detection, then detection speed is improved, but false alarm rate increases
Solution Approach 1:
The patent merges two different detection approaches into a unified system: structure-based detection methods that provide fast initial candidate identification, and statistical verification methods (Adaboost) that reduce false alarms. The combination allows the system to maintain the speed advantages of structure-based methods while incorporating the reliability benefits of statistical verification, thereby reducing false alarm rates without sacrificing detection speed.
Solution Approach 2:
The patent introduces an intermediary verification step using Adaboost statistical classifiers between the structure-based detection and final face identification. This intermediary layer acts as a filter that validates candidates from the structure-based detection, reducing false alarms while maintaining the overall detection speed by only processing candidates rather than all image regions.
3Measurement precision
If conventional face detectors are used, then front face detection is effective, but multi-view face detection performance degrades
Solution Approach 1:
The patent employs Adaboost face detectors that have been trained to recognize faces across multiple views and orientations, making the detection system universal rather than specialized for only front-facing views. These multi-functional detectors can handle various face orientations (front, profile, and intermediate angles) while maintaining detection accuracy, thereby improving adaptability to multi-view scenarios without sacrificing front face detection effectiveness.
4Device complexity
If single orientation edge detectors are used, then detection simplicity is maintained, but orientation independence is lost
Solution Approach 1:
The patent applies edge detectors with different orientation angles (e.g., 0°, 45°, 90°, 135°) to detect edges in various directions. By using asymmetric, multi-oriented edge detectors rather than a single orientation detector, the system achieves orientation independence while maintaining manageable complexity through the systematic application of a limited set of standard orientations that collectively cover all possible face rotations.
Data Source
AI summary
A method for providing orientation independent face detection may include generating multiple mosaic images from an input image in which each of the multiple mosaic images has a different scale, employing a plurality of differently oriented edge detectors to perform edge detection on the multiple mosaic images including combining edges of the multiple mosaic images having the different scales, and performing face detection in regions corresponding to the differently oriented edge detectors based on respective feature maps produced by the differently oriented edge detectors. An apparatus and computer program product corresponding to the method are also provided.