Facial Trait Categorization Using Kernel-Based Codebook Transform

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Current automated methods for categorizing human face images based on visual traits, such as gender, age, and facial features, lack generalization and require separate visual features for each trait, making them inefficient for unconstrained environments and diverse facial characteristics.

Innovation Solution

A method using a kernel-based learning approach with codebook transforms to extract facial traits from point sample data, applying a support vector machine with χ2 kernel, and employing invariant visual features like SIFT and SURF, allowing for generic categorization of human face images without the need for detailed geometry or specific facial region detection.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Measurement precision

If separate visual features are crafted for each visual trait, then the categorization accuracy for that specific trait is improved, but the device complexity and time consumption increase significantly

Engineering Contradiction:
Improvecategorization accuracyVSAvoidsystem complexity
Core Design Contradiction:
Measurement precisionVSDevice complexity

Solution Approach 1:

The patent applies a universal codebook transform framework that can categorize multiple visual traits (gender, age, ethnicity, facial expression, accessories) using the same general process. Instead of creating separate specialized detectors for each trait, the system uses a single unified approach with trait-specific training data, achieving multi-functionality while maintaining accuracy across diverse categorization tasks

Inventive Principle:
Principle #6Universality (Multi-functionality)

Solution Approach 2:

The patent segments the face image into multiple regions (eyes, nose, mouth, cheeks, hair) and extracts visual features from each region independently. These segmented features are then aggregated through the codebook transform to represent the entire face, allowing the system to capture trait-specific information from different facial regions without requiring a completely separate processing pipeline for each trait

Inventive Principle:
Principle #1Segmentation

2Measurement precision

If traditional face detection and recognition methods are used, then face location and identity recognition are improved, but the ability to describe visual appearance traits in unconstrained environments deteriorates

Engineering Contradiction:
Improveface recognition accuracyVSAvoidgeneralization to unconstrained environments
Core Design Contradiction:
Measurement precisionVSAdaptability or versatility

Solution Approach 1:

The patent extracts visual appearance traits from face images by sampling multiple points across the face and extracting features at each point, then aggregating these features through codebook transform. This extraction approach focuses specifically on visual appearance characteristics rather than relying on traditional face detection and recognition pipelines, enabling the system to handle unconstrained environments where faces may be partially occluded, rotated, or under varying lighting conditions

Inventive Principle:
Principle #2Taking out (Extraction)

Solution Approach 2:

The patent changes the representation parameters from traditional face recognition approaches (subspace methods, 3D morphable models) to a codebook transform representation that aggregates visual features into a compact form. This parameter transformation enables better generalization to unconstrained environments by capturing essential visual appearance characteristics in a more robust representation that is less sensitive to pose, lighting, and occlusion variations

Inventive Principle:
Principle #35Parameter changes

3Measurement precision

If detailed geometry and specific facial region detection are performed, then the precision of facial trait identification is improved, but the processing time and computational resources increase

Engineering Contradiction:
Improvefacial trait identification precisionVSAvoidprocessing time
Core Design Contradiction:
Measurement precisionVSLoss of time

Solution Approach 1:

The patent performs preliminary action by pre-defining a codebook of visual features and pre-segmenting the face into regions. During actual processing, the system only needs to extract features at sampled points and aggregate them through the pre-defined codebook transform, rather than performing complex real-time analysis. This preliminary preparation significantly reduces processing time while maintaining precision

Inventive Principle:
Principle #10Preliminary action

Data Source

PatentUS8805030B2Method for automated categorization of human face images based on facial traits
Publication Date: 2014.08.12 QUALCOMM TECHNOLOGIES INC

AI summary

A method for automated categorization of human face images based on facial traits, said method comprising a facial trait extracting phase, comprising the steps of: providing a multitude of images comprising human faces, for each image sampling a multitude of points in said image to obtain point sample data, for each sampled point extracting visual features from said point sample data, for each image assigning said visual features to predefined codewords by applying a codebook transform, for each image extracting facial traits by applying a kernel-based learning method's prediction algorithm to said codewords to establish the probability that a facial trait from a predefined set of facial traits is present in said image, and extract said facial trait for said image if said probability is higher than a predefined threshold.