Abstract：In order to avoid the default of the greedy algorithm to Approximate given function by searching a linear combination of basis functions choosing from a redundant basis function dictionary for the Kernel Matching Pursuits (KMP), we make use of the global optimal searching ability and the locally quickly searching ability of Immune Clonal Selection Algorithm (ICSA) to speed up searching basic function data in function dictionary. And a method for object recognition of Kernel matching pursuits based on immune clonal selection algorithm is presented. This method reduces greatly computer time of the KMP algorithm. The simulation result of the UCI datasets, remote images and Brodatz images show the proposed algorithm can decrease obviously training time leave the classification accuracy almost unchanged, especially for the large size datasets as compared with the standard KMP. The method has higher classification speed and more accurate recognition rate over the matching pursuits based on Genetic Algorithm (GA).
2008, Vol. 30 
