Facial Feature Point Detection Using Reliability Images

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

Existing methods for detecting facial feature points, such as those described in Patent Document 1, fail when less-reliable information is input, leading to inaccurate detection due to improper prior probability calculations and search region settings.

Innovation Solution

An information processing device and method that generates a reliability image using multiple classifiers to assess feature point suitability, calculates candidate positions, and determines conformity to a selected facial shape model, ensuring accurate detection even with less-reliable input by iteratively refining the feature point classification and shape model matching.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Productivity

If prior probability and search region are calculated based on facial feature point candidates, then detection process can be executed, but detection accuracy deteriorates when less-reliable candidates are input

Engineering Contradiction:

Improvedetection process executionVSAvoidfacial feature point detection accuracy

Core Design Contradiction:

ProductivityVSMeasurement precision

Solution Approach 1:

The patent performs preliminary actions by calculating prior probability and search region before the main detection process. However, it refines these calculations iteratively using shape model conformity constraints to ensure accurate detection even when initial candidates are unreliable.

Inventive Principle:

Principle #10Preliminary action

Solution Approach 2:

The patent implements feedback by using the conformity calculation result to adjust and refine the prior probability and search region. The shape model conformity provides feedback information that corrects inaccuracies in the initial detection candidates, enabling the system to maintain high accuracy despite unreliable input.

Inventive Principle:

Principle #23Feedback

2Speed

If single-stage detection is used, then processing speed is maintained, but detection reliability deteriorates under varying conditions

Engineering Contradiction:

Improveprocessing speedVSAvoiddetection reliability

Core Design Contradiction:

SpeedVSReliability

Solution Approach 1:

The patent employs periodic action through iterative refinement processes. The detection algorithm cycles through multiple stages of candidate generation, conformity calculation, and parameter adjustment, periodically improving detection reliability while maintaining acceptable processing speeds through efficient iterative optimization.

Inventive Principle:

Principle #19Periodic action

Data Source

PatentUS9904843B2Information processing device, information processing method, and program

Publication Date: 2018.02.27 NEC CORP

AI summary

An information processing device and method can detect facial feature point positions with high accuracy. The information processing device includes: a facial image input unit receiving a facial image; a facial feature point reliability generation unit generating, based on a plurality of classifiers for computing the suitability of feature points of a face, a reliability image indicating the suitability for each feature point from the facial image; a facial feature point candidate position computation unit obtaining a candidate position of the feature point in the facial image based on the reliability image; and a facial shape model conformity computation unit determining a feature point that satisfies conditions based on a position corresponding to each of the feature points of one facial shape model selected from a plurality of statistically generated facial shape models and the candidate position of the feature point, and calculating a conformity to the facial shape model.