A New Framework of Feature Extraction to Capture Underlying Statistical Information in Image Histogram

Abstract

In recent years, the face recognition application related to Local Binary Pattern (LBP) as features extraction or pattern extraction is actively proposed. The combination of LBP and Gaussian Mixture Model (GMM) is a novel framework where, both are well known as powerful tools. The proposed method is discussing about the new implementation of independent histogram matrix structure from uniform pattern of LBP (LBPu2) LBP8, 1u2 that can be pursue to integrate with GMM to learn the non-parametric data distribution then represented as a convex combination of several normal distributions with respective weight, means and variances. The Experimental used color FERET face image database to evaluate the recognition performance by compared the recognition rate with the existing methods. The proposed framework shown that the average performance of 10-time random test is 98.4%, where the highest performance is 99%.