Impact of illumination and it's analysis on exact face boundary detection
Abstract
The main aim of this project work is to mark the boundaries of human faces in a surveillance video. Since the proposed application is for real-time video streaming performed in an uncontrolled environment, the faces of humans can be disturbed by various factors such as occlusion (error due to motion), overlapping (multiple faces in a frame), illumination errors, etc. The current face detection algorithms like Principle component analysis, Haar Cascade algorithm, Skin color-based algorithms like RGB and HSI are unable to provide exact boundary marking accurately and also not able to overcome these errors. To overcome these challenges an improved algorithm using a combination of shape-based feature extraction and texture-based extraction, that can mark the edges of human faces also removes the illumination effect while the video is streaming with a surveillance camera is proposed. This algorithm will remove the illumination effect and will remove the contours to output the exact boundaries of face.