Gender Recognition from Face Images Using SIFT Descriptors and Trainable Features

Abstract

Nowadays, recognition of gender from facial image became an important problem in computer vision, security, verbal–nonverbal communication and human–computer interaction application. Recognition of gender is a challenging research problem because facial image contains many information such as gender, facial expression, age, ethnic origin in computer-aided applications, based on the facial image quality gender recognition depends. In this paper, a new gender recognition method is proposed that combines both scale-invariant feature transform (SIFT) as domain-specific approach and combination of shifted filter responses (COSFIRE) as trainable features. The proposed method will give better performance of variation in poses, different expressions and changes in illumination condition. This method tested by taking gender face recognition technology (GENDER-FERET) results shows that GENDER-FERET dataset will give better result in various illumination conditions. COSFIRE algorithm which is used for visual recognition problem will work better in given trainable character. The advantage of using SIFT is it will not get affected by changes in scale, blur, rotation, change in illumination and affine transformation.