Abstract

We present here an approach to find the exact position of some feature units related to human face images. We use the horizontal information in most significant bit planes of images to accomplish the task. Finding location of facial feature units is of importance as most human face recognition approaches take it as initial point. The prominent feature units in a face are eyes, nostrils and lips which are usually oriented in horizontal direction and visually significant in face image. The majority of the visually significant data in image can be extracted using higher order bits of that image. Our four step method consists of bit planes processing, separating horizontal information using wavelet transform (WT), binary thresholding and appropriate combination of Dilation and Erosion. The proposed method shows high accuracy in the presence of all real world situations like various gestures, illumination variations, closed eyes, and eyes with glasses.

Key words: Bit planes, feature localization, facial features, face recognition and wavelet transform.