A Novel Technique for Human Face Detection Based on Color Segmentation and Template Matching

This paper presents a novel face detection algorithm. Face Detection is the process of locating faces in the image. Our aim is to locate all faces in the image irrespective of various challenges usually faced by researchers. Image including faces is the further key to intelligent vision-based human computer interaction & research attempts in face processing including face recognition, face motion tracking, pose estimation, and feature recognition. Various approaches have been employed to detect faces in images. There are various challenges associated with the different techniques. The challenges usually accustomed by researchers are the facial gestures, texture, pose and lighting conditions while taking an image. Procedure used by us does not impinge on the cited challenges. The technique used for identifying the face in an image includes skin colour segmentation to reject the non-skin region, removal of noise and clutters in the segmented image through a series of morphological operations, masking the morphological operated image over the gray scale image, rejection of non-face region like hands, arms, etc based on geometry and Euler number, convolving our last output image with the templates to get final gray image containing only faces. Skin colour segmentation is done by plotting the histogram of HSV colour image and thus separating the skin region from the image. Non-face skin region is then rejected by convolving the morphological output with the templates that are representatives of human faces. Various efficient face detection techniques uses neural network scanning of test image and executes exhaustive search in every part of the image, so increases the search time and complexity. With this technique we have overcome the challenge faced by the previous techniques, i.e. time factor. In experiments on images, our algorithm has achieved high detection rates and less false detection due to the application of the new algorithm.