Document Type
Conference Item
Publication Date
12-1-2011
Abstract
In this paper, we present a neural networkbased method to detect frontal faces in grayscale images under unconstrained scene conditions such as the presence of complex background and uncontrolled illumination. The system is composed of two stages: threshold-based segmentation and neural network-based classifier. Image segmentation using thresholding is used to reduce the search space. Artificial neural network classifier would then be applied only to regions of the image which are marked as candidate face regions. The ANN classification phase crops small windows of an image, and decides whether each window contains a face. Partial face template is used instead of the whole face to make training process easier. To minimize the probability of misrecognition, texture descriptors such as mean, standard deviation, smoothness and X-Y-Relieves are measured and entered besides the image as input data to form solid feature vector. The ANN training phase is designed to be general with minimum customization and to output the presence or absence of a face (i.e. face or non-face). In this work, partial face template is used instead of the whole face. Aligning faces is done using only one point that is “face center”.
Keywords
Thresholding, Image segmentation, Artificial neural network, Texture analysis, Face detection
Divisions
fsktm
Event Title
International Arab Conference on Information Technology
Event Location
Riyadh, Arab Saudi
Event Dates
11-14 Dec 2011
Event Type
conference