Face Detection Using Angle-Specific Characteristic Data
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing image capturing systems face challenges in rapidly detecting face areas due to the burden of successive face detection and image processing, which can be adversely affected by changes in camera angle, especially when used in handheld devices.
Innovation Solution
A digital camera system that prestores characteristic data of face areas at different camera rotation angles and uses this data to rapidly detect and process face areas in live-view images, reducing the load on subsequent full-size image processing by selecting the most relevant data based on camera angle and priority.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If successive face detection and image processing are performed on all captured images, then face detection accuracy is improved, but processing time and computational burden increase significantly
Solution Approach 1:
The patent applies partial action by performing exhaustive face detection only on selected images based on predetermined conditions (e.g., detection of skin color areas, specific image quality criteria), while using simplified or no detection on other images. This selective approach maintains high detection accuracy for important images while reducing overall processing time and computational burden across the entire image set.
2Adaptability or versatility
If face detection is performed on images captured at varying camera angles, then adaptability to different shooting scenarios is improved, but detection accuracy deteriorates due to angle-induced distortions
Solution Approach 1:
The patent applies local quality by storing and utilizing characteristic data specific to different camera rotation angles. The system detects the current camera angle and selectively accesses pre-stored characteristic data corresponding to that specific angle, allowing accurate face detection adapted to local angular conditions rather than using a universal detection model that would compromise accuracy.
Solution Approach 2:
The patent applies preliminary action by pre-storing characteristic data of face areas at multiple predetermined camera rotation angles before actual face detection occurs. This advance preparation enables the system to quickly retrieve angle-specific characteristics during detection, improving both adaptability to varying angles and detection accuracy without real-time computational overhead.
3Measurement precision
If full-resolution images are processed for face detection, then detection precision is improved, but processing load and time consumption increase
Solution Approach 1:
The patent applies segmentation by dividing the face detection process into multiple stages: initial detection on down-sampled low-resolution images to identify candidate face areas, followed by refined detection only on those candidate regions using full-resolution image data. This segmented approach maintains high detection precision for final results while significantly reducing overall processing load by limiting full-resolution analysis to small candidate regions rather than entire images.
Data Source
Figure 1
Figure 2
Figure 3
AI summary
In an image capture mode of a camera, a face area is detected from each of live-view images of a subject captured periodically (step S2). Information on the detected face area is stored in a detection result storage area (131) (step S4). When a shutter button is fully depressed (YES in step S5), a full-size image is acquired (step S6). Then, it is determined whether face area information is stored in the area 131 (step S7). If so (YES in step S7), information on an angle through which the camera is rotated to obtain a face area with the highest selection priority is selected from the face area information stored (step S8). Then, the face area detecting process is performed on the full-size image, using characteristic data on the face area involving the selected angle information (step S9) .