Face recognition using truncated transform domain feature extraction

Abstract EN

Face Recognition (FR) under varying pose is challenging and exacting pose invariant features is an effective approach to solve this problem.

In this paper, we propose a novel Truncated Transform Domain Feature Extractor (TTDFE) to improve the performance of the FR system.

TTDFE involves a unique combination of Symlet - 4 DWT, 2D - DCT, followed by a novel truncation process.

The truncation process extracts higher amplitude coefficients from the Discrete Cosine Transform (DCT) matrix.

An optimal truncation point is estimated, which is inspired by a relationship developed between the image dimensions and the positions of DCT amplitude peaks.

TTDFE is used for efficient feature extraction and a Binary Particle Swarm Optimization - based feature selection algorithm is used to search the feature space for the optimal feature subset.

Experimental results, obtained by applying the proposed algorithm on 5 benchmark face databases with large pose variations, namely FERET, UMIST, FEI, PHPD and IFD, show that the proposed system outperforms other FR systems.

A significant increase in the Recognition Rate and a substantial reduction in the number of features selected, are observed.