Halftone Based Face Recognition Using SVM

We propose a face recognition method based on halftone binary image using SVM classifier. In this method, a training set and a testing set of halftone images are created from a face database of gray images. Then, features are extracted from halftone images and a multi-class SVM model is created. To extract features from a halftone image, the image is divided into non-overlapping regions of equal size. Each region is processed to give a feature value corresponding to the region. This reduces the size of feature vector depending on the size of region considered for a feature. Four different types of features can be generated depending on how the processing of the pixels in each region is done to generate a feature. Recognition rate is computed for each of the four different types of features. Three different types of features give comparatively higher recognition rate for different window sizes. The method has been tested on AT&T face database using different feature types and window sizes. In one of feature types, it gives recognition rate of 95% which much higher than recognition rate 91.25% when using with HoG features on the same face database.