Facial Contour Recognition Method and Device Using Grayscale Variance

Rapid Facial Contour Recognition Using Patch-Based Grayscale Variance Comparison

This technology is a facial contour recognition method and device that projects a reference shape onto an input image to create a frontal face image, forms patches for each contour landmark, and calibrates these patches by calculating the grayscale variance of each.

Existing shape model-based facial contour detection often applied models even when initial landmark positions were inaccurate, making it difficult to assess result reliability beforehand. This led to increased computational load and misaligned contours.

This technology generates a frontal face using bilinear interpolation and verifies result reliability before applying the shape model by resetting landmarks based on a comparison of patch grayscale variance proportional to the overall shape size. Applicable to access security, smartphone facial recognition, and video-based expression analysis, it enables fast facial recognition with reduced computational requirements and fewer contour detection errors.

‍

‍

Key Features:

• Projecting a reference shape onto the acquired image and excluding non-facial areas to generate a frontal face image
• Extracting the contours of the frontal face image, using them as landmarks, and forming patches on the contours proportional to the overall shape size
• Calculating the grayscale variance for each patch based on the grayscale values within the landmark-centered patches
• Calibrating the patches by comparing the grayscale variance of the patches to reset landmarks and reform the patches

‍