Fast Feature Extraction Approach for Multi-Dimension Feature Space Problems

Recently, we proposed a fast feature extraction approach denoted FSOM utilizes Self Organizing Map (SOM). FSOM [1] overcomes the slowness of traditional SOM search algorithm. We investigated the superiority of the new approach using two lip reading data sets which require a limited feature space as the experiments in [1] showed. In this paper, we continue FSOM investigation but using an RGB face recognition database across different poses and different lighting conditions. We believe that such data sets require multi-dimensional feature space to extract the information included in the original data in an effective way especially if you have a big number of classes. Again, we show here how is FSOM reduces the feature extraction time of traditional SOM drastically while preserving same SOM’s qualities.

• Type: Conference
• Year: 2006
• Where: ICPR
• Authors: Alaa El. Sagheer, Naoyuki Tsuruta, Rin-ichiro Taniguchi, Daisaku Arita, Sakashi Maeda
• Keywords: Computer Vision, Feature Extraction, ICPR 2006, Traditional Som, Traditional Som Search