Adaptive Skin Tone Detection for Low-Cost Mobile Face Recognition
Solution Overview
Problem
Existing face detection methods for mobile devices require either dedicated hardware or strong CPU capabilities, making them unsuitable for low-cost mobile platforms, and struggle with dynamic settings like auto-white balance, which affects skin tone detection accuracy.
Innovation Solution
A method that dynamically defines an extended range of color component values in the CbCr color space, scans intervals to detect skin tones, and selects candidate limited ranges based on the number of detected faces, reducing computational complexity and false detections, while adapting to changing conditions.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If dedicated hardware or strong CPU capabilities are used for face detection, then detection accuracy and real-time performance are improved, but device cost and complexity increase
Solution Approach 1:
The patent dynamically adjusts color space ranges and intervals based on image characteristics and detection results, transforming fixed parameter detection into adaptive parameter detection. This allows accurate face detection without requiring dedicated hardware by optimizing software algorithms to adapt to varying image conditions.
Solution Approach 2:
The patent implements dynamic range adjustment where the color component ranges are not fixed but are adapted based on the specific image being analyzed and previous detection results. This dynamic approach enables the system to maintain high detection accuracy across different lighting and skin tone conditions without needing hardware adjustments.
2Productivity
If a fixed color range is used for skin tone detection, then computational complexity is reduced, but detection accuracy decreases under varying lighting conditions
Solution Approach 1:
The patent transforms the static color range into a dynamic range that adapts to different lighting conditions and image characteristics. The system determines optimal color component ranges based on the specific image being processed, maintaining both processing efficiency and detection accuracy across varying conditions.
Solution Approach 2:
The patent performs preliminary analysis of the image to determine appropriate color ranges before conducting full face detection. This preliminary action prepares the optimal detection parameters in advance, allowing the main detection process to proceed efficiently with pre-determined ranges that are already suited to the image characteristics.
3Measurement precision
If the entire extended color range is scanned for skin tone detection, then detection coverage is improved, but computational complexity and processing time increase
Solution Approach 1:
The patent divides the extended color range into multiple intervals and processes them in a structured sequence. By segmenting the search space and using interval-based detection with early termination criteria, the system achieves comprehensive coverage without requiring exhaustive scanning of the entire color space.
Solution Approach 2:
The patent implements interval-based detection where not all possible color ranges are fully explored. Instead, detection is performed on selected intervals that are most likely to contain skin tones, with the ability to terminate early when sufficient detections are achieved, thus avoiding unnecessary processing of the entire extended range.
4Productivity
If multiple color intervals are scanned simultaneously in parallel, then processing efficiency is improved, but resource requirements and algorithm complexity increase
Solution Approach 1:
The patent segments the detection task into independent interval processing units that can be executed in parallel. Each interval is processed independently with its own detection logic, allowing for parallel implementation without requiring complex coordination between processing threads, thus maintaining relatively simple algorithm structure while achieving improved efficiency.
Data Source
Figure 1~4
Figure 3
Figure 5~6
AI summary
There is described a method for face detection in an image comprising a plurality of picture elements, each picture element having a plurality of colour components in a predetermined colour space. The method comprises determining (60) an extended range for colour component values, in the colour space, in which a skin tone area is likely to be detected; defining (62) a plurality of intervals for the colour component values, in the colour space, covering at least part of the extended range; scanning (64) each of the plurality of intervals to detect a skin tone area; and, if at least one skin tone area is detected in at least one interval, then selecting (60) the intervals in which at least one skin tone area is detected; defining (68) candidate limited ranges for colour component values, in the colour space, from the selected intervals; performing (70) a face detection on at least one skin tone area in at least some of the candidate limited ranges and selecting a chosen candidate limited range based on face locations detected during the face detection.