Adaptive Face Feature Extraction via Landmark Angles
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing face recognition systems face challenges in accurately recognizing faces with significant pose changes and varying distances, leading to reduced recognition rates due to normalization processes and feature extraction inefficiencies.
Innovation Solution
A method that detects landmarks in an input image, calculates angles between them, and determines a feature extraction scheme based on these angles to extract features from specific areas in a predetermined order, adapting to pose changes by selecting appropriate feature extraction schemes for slight or great pose variations.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If normalization process is performed to transform face to front face and adjust size, then face recognition can be performed, but recognition accuracy decreases when pose changes are substantial
Solution Approach 1:
The patent applies dynamics by making the feature extraction process adaptive to pose changes. Instead of using a fixed normalization process, the system dynamically selects extraction schemes based on detected pose angles. The feature extraction is adjusted in real-time according to the user's pose, allowing the system to maintain accuracy across various orientations without requiring substantial pose transformation.
Solution Approach 2:
The patent changes parameters by using pose-dependent extraction schemes. Different feature extraction parameters and target areas are selected based on the detected pose angles. When pose angles exceed threshold values, the system switches to alternative extraction schemes that are optimized for non-frontal views, thereby maintaining recognition accuracy across different pose conditions.
2Productivity
If feature extraction is performed from normalized facial image, then recognition can be performed, but feature extraction efficiency decreases due to pose variations
Solution Approach 1:
The patent applies preliminary action by detecting landmarks and calculating pose angles before performing feature extraction. This preliminary pose assessment allows the system to pre-select the most appropriate extraction scheme, avoiding unnecessary processing steps and ensuring that the correct features are extracted efficiently for the given pose condition.
Solution Approach 2:
The patent applies local quality by determining specific target areas for feature extraction based on pose conditions. Instead of extracting features uniformly from the entire face, the system identifies and extracts features from relevant local regions that are most informative for the current pose, thereby improving both efficiency and accuracy.
3Adaptability or versatility
If normalization process transforms size of facial image, then face recognition can be performed at distance, but recognition rate decreases due to loss of original feature characteristics
Solution Approach 1:
The patent applies segmentation by dividing the face into multiple landmarks and analyzing their geometric relationships. This landmark-based approach captures essential facial structure information that is scale-invariant, allowing the system to recognize faces at varying distances without relying solely on image size normalization, thus preserving feature characteristics across different scales.
Data Source
AI summary
At least one example embodiment discloses a method of extracting a feature from an input image. The method may include detecting landmarks from the input image, detecting physical characteristics between the landmarks based on the landmarks, determining a target area of the input image from which at least one feature is to be extracted and an order of extracting the feature from the target area based on the physical characteristics and extracting the feature based on the determining.