Adaptive Facial Expression Recognition Thresholds
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing facial expression recognition technologies face challenges in accurately distinguishing between facial expressions with similar shapes, such as eyes and mouths, leading to recognition errors, especially in cases where cheek muscles move or eyes are half-closed, resulting in incorrect classification of smiling, crying, or embarrassed faces due to individual differences and wasteful processing.
Innovation Solution
A facial expression recognition apparatus that includes image input, face detection, calculation, updating, and determination means, which calculates facial expression evaluation values, updates relationships between these values and thresholds, and determines facial expressions based on priority weights to differentiate between similar expressions.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If facial expression recognition is performed based on maximum value of evaluation values from multiple discriminators, then facial expressions can be determined, but recognition errors occur when different expressions have similar shapes (e.g., smiling vs. blinking)
Solution Approach 1:
The patent changes the parameter of threshold values dynamically based on the detected facial expression type. When a smiling expression is detected, the threshold for eye-closed recognition is raised to prevent false positives. This parameter adjustment resolves the contradiction by adapting the recognition criteria to the current facial state, improving both accuracy and reliability.
Solution Approach 2:
The system dynamically adjusts recognition thresholds based on the detected facial expression rather than using fixed thresholds. The threshold for determining eye-closed expressions changes according to the overall facial expression context, making the recognition system adaptive and reducing errors between similar expressions like smiling and blinking.
2Measurement precision
If calculation for facial expression evaluation values is performed uniformly for each expression, then comprehensive evaluation is achieved, but wasteful processing occurs when eyes are obviously closed
Solution Approach 1:
The system performs preliminary detection of obvious eye-closed states before conducting full facial expression evaluation. When eyes are clearly closed, the system quickly identifies this state and skips comprehensive calculation for other expressions, preventing wasteful processing while maintaining evaluation completeness for relevant expressions.
Solution Approach 2:
The patent extracts and handles the eye-closed state as a separate, priority condition. By detecting obvious eye-closed states first and treating them as a distinct case, the system removes the need for uniform comprehensive calculation across all expressions, improving processing efficiency without sacrificing evaluation accuracy for the detected state.
3Reliability
If image sensing is performed based on detected facial expressions, then optimal images are captured, but incorrect images are captured due to recognition errors from similar shapes
Solution Approach 1:
The patent adjusts the recognition thresholds dynamically based on the detected facial expression before triggering image capture. When a smiling expression is detected, the threshold for eye-closed recognition is increased, preventing false capture of blinking images as smiling images. This parameter change ensures that only accurately recognized expressions trigger image sensing.
Data Source
AI summary
A facial expression recognition apparatus (10) detects a face image of a person from an input image, calculates a facial expression evaluation value corresponding to each facial expression from the detected face image, updates, based on the face image, the relationship between the calculated facial expression evaluation value and a threshold for determining a facial expression set for the facial expression evaluation value, and determines the facial expression of the face image based on the updated relationship between the facial expression evaluation value and the threshold for determining a facial expression.