Face Direction Estimation and Detection via Manifold Learning

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

Existing face direction estimation and facial detection technologies face challenges in simultaneously learning face information and face direction information without significant cost, especially when these details are not provided for all images in advance.

Innovation Solution

An image processing system that includes units for identifying known and unknown face directions, converting face direction information into manifold positions, estimating suitable positions on the manifold, identifying facial images, and updating parameters based on distances between manifold and image positions to perform face direction estimation and facial detection with high accuracy.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Reliability

If face direction estimation and facial detection are performed as independent technologies with pre-prepared image groups, then each processing can achieve reasonable accuracy, but the system complexity increases and cannot be performed simultaneously with high accuracy

Engineering Contradiction:

Improveaccuracy of face direction estimation and facial detectionVSAvoidsystem complexity

Core Design Contradiction:

ReliabilityVSDevice complexity

Solution Approach 1:

The patent merges face direction estimation processing and facial detection processing into a single integrated system that performs both functions simultaneously. The learning device combines multiple learning data groups (frontal face images, inclined face images, and non-face images) into a unified learning framework, allowing both processing tasks to be executed together with high accuracy without requiring separate independent systems.

Inventive Principle:

Principle #5Merging (Combining)

Solution Approach 2:

The patent creates a universal learning system that handles multiple functions: facial detection, face direction estimation, and non-face identification. By using a single learning device that processes all three types of image data together, the system achieves multi-functionality where one system performs what previously required multiple separate processing chains.

Inventive Principle:

Principle #6Universality (Multi-functionality)

2Reliability

If face information and face direction information are provided for all images in advance, then simultaneous learning achieves high accuracy, but the cost and time for manual labeling increases significantly

Engineering Contradiction:

Improveaccuracy of simultaneous face detection and direction estimationVSAvoidtime for manual labeling

Core Design Contradiction:

ReliabilityVSLoss of time

Solution Approach 1:

The patent segments the learning data into three distinct groups: frontal face images with direction information, inclined face images with direction information, and non-face images. This segmentation allows the system to learn different aspects separately while maintaining overall integration, reducing the burden of complete manual labeling for all images while preserving accuracy.

Inventive Principle:

Principle #1Segmentation

Solution Approach 2:

The patent applies partial labeling by providing face direction information for only certain image groups (frontal and inclined face images) rather than all images. Non-face images and some facial images use simpler labeling. This partial action approach maintains learning accuracy while significantly reducing the time and cost of manual labeling compared to complete labeling of all images.

Inventive Principle:

Principle #16Partial or excessive action

3Ease of operation

If face direction estimation premises that facial detection has been performed, then the processing flow is simplified, but the system cannot handle images where face direction is unknown

Engineering Contradiction:

Improveprocessing flow simplicityVSAvoidability to handle unknown face directions

Core Design Contradiction:

Ease of operationVSAdaptability or versatility

Solution Approach 1:

The patent implements a dynamic processing framework where the system can adapt its operation mode based on input requirements. The learning device is configured to perform either sequential processing (facial detection followed by face direction estimation) when face direction is known, or simultaneous processing when face direction needs to be determined together with facial detection, providing flexibility to handle various scenarios.

Inventive Principle:

Principle #15Dynamics

Solution Approach 2:

The patent prepares multiple types of learning data groups in advance (frontal face images, inclined face images, non-face images) with appropriate labeling information. This preliminary preparation of diverse data groups enables the system to handle various cases including unknown face directions without requiring complex runtime decisions, as the learning framework is already configured to process all types.

Inventive Principle:

Principle #10Preliminary action

Data Source

PatentUS8971613B2Image processing learning device, image processing learning method, and image processing learning program

Publication Date: 2015.03.03 NEC CORP

AI summary

Disclosed is an image processing learning device with which face direction estimation processing and face detection processing can be executed simultaneously and with high precision without incurring significant costs. The image processing learning device comprises: a face direction identification unit, a position estimation unit, a face identification unit, a first update quantity calculation unit, a second update quantity calculation unit, and a parameter update unit.