Texture Classification Based on Coevolution Approach in Multiwavelet Feature Space
To test the effectiveness of multiwavelets in texture classification with respect to scalar Daubechies wavelets, we study the evolutionary-based algorithm to evaluate the classification performance of each subset of selected feature. The approach creates two populations that have interdependent evolutions corresponding to inter and intra distance measure, respectively. With the proposed fitness function composed of the individuals in competition, the evolution of the distinct populations is performed simultaneously through a coevolutionary process and selects frequency channel features of greater discriminatory power. Consistently better performance of the experiments suggests that the multiwavelet transform features may contain more texture information for classification than the scalar wavelet transform features. Classification performance comparisons using a set of twelve Brodatz textured images and wavelet packet decompositions with the novel packet-tree feature selection support this conclusion.