Automatic Face Segmentation Using Adaptively Regularized Kernel-Based Fuzzy Clustering Means With Level Set Algorithm

Abstract

In this research, a new level set-based segmentation algorithm was proposed for human face segmentation. At first, the human facial images were collected from face semantic segmentation (FASSEG) dataset. After collecting the images, pre-processing was accomplished by utilizing contrast limited adaptive histogram equalization (CLAHE). The undertaken methodology effectively improves the quality of facial images by removing the unwanted noise. Then, segmentation was done by using adaptively regularized kernel-based fuzzy clustering means (ARKFCM) clustering with level set, which was a high-level machine learning algorithm for localizing the face parts in complex template. Simulation outcome shows that the proposed segmentation algorithm effectively segments the facial parts in light of precision, recall, Jaccard coefficient, dice coefficient, accuracy, and miss rate. The proposed segmentation algorithm enhanced the segmentation accuracy in face segmentation up to 4.5% compared to the existing methodology (pixel wise segmentation).