Hierarchical dictionary-based technique for face recognition system

EasyChair Preprint 1587

Abstract

Face recognition still the most studied topic in the pattern recognition research field. This is probably due to the multiple useful applications which can be developed for important domains like HMC, Forensic, Security, Entertainment etc. Such deployed research efforts produced a huge number of methods, techniques and algorithms with different characteristics according to their simplicity, efficiency, robustness and speed. In the present work, we investigate the performances of a simplified technique using a hierarchical classification scheme based on a constructed multi parts dictionary of elementary blocs obtained by applying sequential classifier to the whole set of features of the working database. The elementary features of the different parts of the constructed dictionary were obtained using the well-known and simplest and effective way to depict the similarities between two compared patterns; named cross-correlation operator applied to the original images and their transformed images known as integral images. Hierarchical clasification scheme is used to overcome the fact that this operator has high consumption time cost. The proposed strategy was implemented and tested on the images of the well known ORL and YALE database sets. Practical results demonstrate largely recognizable efficiency and speed characteristics.

Keyphrases: Integral Image, K-means classifier, cross-correlation, face recognition, hierarchical structure, pattern recognition, processed image

@booklet{EasyChair:1587,
  author    = {Mohammed Saaidia and Messaoud Ramdani},
  title     = {Hierarchical dictionary-based technique for face recognition system},
  howpublished = {EasyChair Preprint 1587},
  year      = {EasyChair, 2019}}