Component-based Gender Identification from Facial Images

Abstract

As one of the most flourishing applications of image analysis and understanding, face identification has gained significant attention, especially in the past several decades. In the field of image processing, face is one of the most important biometric traits and is becoming more popular for many application purposes now a days. This is a very important field of image processing because of its applications in many areas like security, monitoring, surveillance, commercial profiling, and human-computer interaction. Most previous researchers have been using whole face to classify gender. In this study many techniques for gender classification proposed and the experimental results have shown that the proposed techniques have high accuracy rate, fast computational time besides reduction in the processing time. The study applied in many situations (whole face and components of the face) of gender from facial images using different datasets (FERET, ESSEX and UOFG). Then two types of face detecting algorithms, namely viola & jones and Discriminative Response Map Fitting DRMF model, are applied. Many feature extraction techniques are also applied; global Discrete Cosine Transform (DCT), Block based Discrete Cosine Transform (BBDCT) and hybrid DCT Discrete wavelet Transform DWT) and Local Binary Pattern (LBP), Local Directional LDP), Local Ternary pattern (LTP) and proposed Dynamic Local Ternary Pattern (DLTP) then CNN), then KNN and SVM are used in the last step to classify the images. The experimental results show that the performance of recognition (classification) does not depend only on feature extraction approach but also on the other steps of the recognition process such as the pre-processing stages and the classification algorithm. Moreover, it has been proved that the whole face is not required for gender identification using facial images. This has confirmed that the proposed Dynamic local ternary pattern (DLTP) is an accurate and efficient feature extraction technique for gender identification. Finally, ResNet-50 is applied with CNN to extract the feature of facial images and obtained features in the last fully connected layer which are used as inputs to SVM classifier to produce the final classification result. The study shows that when we used FERET datasets and SVM classifier in the whole face, the accuracy is 95%; but when we used components of the face the accuracy is 98.55%.