Face Recognition in Compressed Domain by Applying Canonical Correlation Analysis Based Feature Vector Optimization and Neural Network Matching

Free registration required

Executive Summary

This paper presents an efficient approach for performing face recognition in compressed domain by applying Canonical Correlation Analysis based feature vector optimization. CCA is a dominant method for multivariate analysis and therefore a powerful method of feature projection based on CCA is proposed for compressed facial images. A major advantage of the proposed approach is the fact that face recognition systems can straightly work with JPEG2000 compressed images as it uses entropy points as input to the new face recognition system based on CCA.

  • Format: PDF
  • Size: 303.65 KB