Facial Feature Extraction Using Probabilistic Skin-Component Segmentation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current facial recognition technologies face challenges in accurately extracting facial features from images, particularly due to the interference of skin regions which can decrease the characteristic quality of facial components like eyes, nose, and lips, and are not robust against variations in illumination and skin color.
Innovation Solution
A method and apparatus that extract facial landmarks, generate probabilistic models for skin and facial component regions, and use these models to isolate the facial component region from the skin region, allowing for the extraction of contour-level, region-level, or pixel-level facial features using techniques like chord angle classification and polar shape matrices, thereby enhancing feature extraction robustness.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of manufacture
If facial features are extracted from the entire face region, then the extraction process is simple, but the skin region interferes with and decreases the characteristic quality of facial components
Solution Approach 1:
The face region is segmented into skin region and facial component region based on probabilistic models. The system divides the complex task of extracting facial features from the entire face into separate processing stages: first classifying pixels into skin vs. facial component regions, then extracting features only from the isolated facial component region, thereby eliminating skin region interference while maintaining extraction simplicity
Solution Approach 2:
The skin region is extracted and removed from the facial component region through probabilistic classification. By identifying and separating the skin region pixels from the facial component pixels based on learned probabilistic models, the system isolates only the relevant facial component information for feature extraction, eliminating the harmful interference of skin regions
2Productivity
If traditional facial feature extraction methods are used, then the process is fast, but the methods are not robust against variations in illumination and skin color
Solution Approach 1:
The system changes the parameter space from raw pixel values to probabilistic class probabilities. Instead of using direct pixel intensity values that are sensitive to illumination and skin color, the system transforms pixel data into probabilistic classifications of skin vs. facial component regions, which are invariant to these variations, thereby achieving robustness while maintaining processing speed
Solution Approach 2:
Probabilistic models serve as an intermediary between the input image and the facial feature extraction. These models first process the image data by classifying each pixel into skin or facial component regions based on probabilistic likelihoods, creating a transformed representation that is robust to illumination and skin color variations before feature extraction occurs
Data Source
AI summary
A method and an apparatus for extracting a facial feature and a method and an apparatus for recognizing a face are provided, in which the apparatus for extracting a facial feature may extract facial landmarks from a current input image, sample a skin region and a facial component region based on the extracted facial landmarks, generate a probabilistic model associated with the sampled skin region and the facial component region, extract the facial component region from a face region included in the input image using the generated probabilistic model, and extract facial feature information from the extracted facial component region.