Facial Recognition Using Skin Mask and Correlation Filter

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Facial recognition systems face challenges in accurately identifying individuals from non-occluded images when the face is obstructed by objects like scarves, hats, or sunglasses, as existing methods struggle to effectively utilize occluded images for identification.

Innovation Solution

The method involves generating a skin mask and correlation filter from an occluded image, which is applied to non-occluded input images to determine the likelihood of matching, using techniques like Fast Fourier Transform and Optimal Trade-off Maximum Average Correlation Height algorithms to enhance recognition accuracy.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Adaptability or versatility

If facial recognition systems use occluded images for identification, then the system can identify individuals even when faces are obstructed, but the recognition accuracy decreases compared to non-occluded images

Engineering Contradiction:
Improveability to identify occluded facesVSAvoidrecognition accuracy
Core Design Contradiction:
Adaptability or versatilityVSMeasurement precision

Solution Approach 1:

The system performs preliminary processing on occluded images to generate skin masks that identify skin regions before recognition. By pre-segmenting the image to isolate skin areas and remove occlusions, the system prepares optimized input for the correlation filter, thereby maintaining high recognition accuracy even with occluded faces

Inventive Principle:
Principle #10Preliminary action

Solution Approach 2:

The invention extracts and removes occluded areas from the image by generating a skin mask that identifies and isolates skin regions. This extraction process separates the useful skin information from harmful occlusion elements, allowing the correlation filter to work only with relevant facial features and achieve accurate recognition despite occlusions

Inventive Principle:
Principle #2Taking out (Extraction)

2Productivity

If the system processes multiple non-occluded images to identify individuals, then the identification coverage increases, but the computation time and processing complexity increase

Engineering Contradiction:
Improveidentification coverageVSAvoidcomputation time
Core Design Contradiction:
ProductivityVSLoss of time

Solution Approach 1:

The system pre-generates skin masks and correlation filters from occluded images before actual recognition tasks. This preliminary preparation creates optimized recognition models that can be quickly applied to multiple non-occluded images, enabling high-volume processing without proportionally increasing computation time

Inventive Principle:
Principle #10Preliminary action

Solution Approach 2:

The invention creates a correlation filter as a computational copy or representation of the occluded face's skin pattern. This filter can be rapidly applied to multiple test images without re-processing the original occluded image, enabling efficient batch processing of numerous non-occluded images against a single occluded reference

Inventive Principle:
Principle #26Copying

Data Source

PatentUS9704025B2Identifying non-occluded faces by learning from partially occluded faces
Publication Date: 2017.07.11 KING ABDULAZIZ CITY FOR SCIENCE AND TECHNOLOGY
  • US9704025B2 patent drawing
  • US9704025B2 patent drawing
  • US9704025B2 patent drawing

AI summary

Systems and methods of facial recognition of non-occluded or non-obstructed faces is provided. The method includes generating a skin mask based on an occluded image of an individual and removing occluded areas for the occluded image using the skin mask so that skin areas remain. The method further includes generating a correlation filter based on the occluded image. The method further includes applying the skin mask and correlation filter to a non-occluded input image of an individual. The method further includes generating a correlation output based on applying the skin mask and correlation filter to the non-occluded input image. The correlation output may indicate a likelihood that the second individual in the non-occluded input image is of the first individual in the occluded image.