SOM competition for complex image scene with variant object positions

T. Tobely,Yuichiro Yoshiki,Ryuichi Tsuda,Naoyuki Tsuruta,M. Amamiya

Abstract:*Department of Intelligent Systems, Graduate School of Information Scienceand Electrical Engineering, Kyushu University, JAPAN**Department of Electronics Engineering, Graduate School of ElectronicsEngineering, Fukuoka University, JAPANAbstract. In this paper, a new SOM competition algorithm is proposed forimage recognition applications. The competition in this algorithm depends on asubset of most discriminate weights of the network codebooks. This indeed canreduce the required recognition time of one image. In addition, the competitionis applied on the pixels corresponding to the object graylevels only, this allowsrecognizing complex images with different lighting conditions. Furthermore, toallow shift variations in the position of the input object, window-basedcompetition is proposed. Where, different subset windows are selected from theinput image, then the competition is applied between each window and windowof the same size in the center of the codebook of all feature map neurons. Theexperimental results of the new algorithm showed good performance inrecognizing gestures of complex images with variant object position while thenormal SOM competition algorithm completely failed.