Performance Evaluation of Various Feature Extraction Techniques with Special Reference to Hand Gesture Recognition

Extraction of significant and dominant features from possibly large set of database is a crucial task. The Performance of feature extraction technique depends on the dimensions of generated features and reconstruction. In this paper we provide a comparative study of different feature extraction techniques like Discrete Cosine Transform (DCT), Discrete Wavelet Transform (DWT), Principle Component Analysis (PCA), Local Binary Pattern (LBP), DCT+Gabor, DWT+ Gabor etc. and each technique is compared with each other based on Sebastian Marcel static hand postures database[24] consisting of six postures. We have used Neural Network to compare the performances of feature extraction techniques based on Recognition Accuracy (RA), False Acceptance Rate (FAR), False Recognition Rate (FRR) and also dimensions. We found that fusion of LBP and Gabor, DWT and Gabor provides good results.