A Novel Approach of Multi-Objective Based Feature Selection Method for Face Recognition

In Face Recognition (FR), image representation is a longstanding practice. FR is complex due to the rich context and a large number of details in photographs. As a result, the image features must be correctly extracted and understood. It’s challenging to come up with decent picture descriptors. One of the significant issues in pattern recognition is the dimensionality of extracted feature vectors; hence feature selection becomes a necessity. Feature selection is an essential and necessary step for use in bio-metrics as bio-metrics tends to degrade over time. Thus to solve this problem, a multi-objective feature selection method has been proposed. The prime objectives of the feature selection method were to minimize the variance within the same group so that it is homogeneous while maximizing the variance between the groups to be well separated. Feature selection is coupled with the three widely used Feature Extraction(FE) methods Principle Component Analysis (PCA), Wavelet Transform (WT), and Local Binary Pattern (LBP). A maximum of 95.03 percent recognition rate was achieved when wavelet FE was used combined with the hybrid FS method. In contrast, a 94.21 percent recognition rate was achieved when the feature selection method was not used.