Video-Based Face Recognition Using HMM and Details Horizontal Orientation by 2D-DWT: Application to VIDTIMIT Database

Abstract

This paper proposes an approach based on two-dimensional discrete wavelet transformation for face recognition, based on low frequency horizontal and vertical high-frequency for a sampling horizontal of faces. This paper also suggests a model for the face recognition under large illumination variations in the videos. It shouts as well the robustness of the system with respect to the fitting variation and the luminance by the use of the histogram remapping technique in the preprocessing step. The face recognition method is based on Hidden Markov Models (HMMs) with the use of a top-to-bottom architecture. Many of these HMM’s built for each individual with robust characteristics in wide variation in illumination. Facial features are retrieved via the use of two-dimensional discrete wavelet transform in the parameterization phase. For faces images, the wavelet of Haar representation differentiates several spatial orientations. We study the application of this representation to compared data in faces recognition. This method has an accuracy of 95%. In fact, it gives the best recognition percentage if compared to any other method reported so far on VIDTIMIT, video database.