Application of hierarchical self-organizing mapping to invariant recognition of color-texture images

In this paper, we present a hierarchical self-organizing map applying to scaling and rotation invariant recognition of a 256/spl times/256-pixel color-texture image. Since Kohonen's self-organizing mapping is not embedded with the invariant ability, some learning modifications are added in rotation and scaling invariant self-organizing map (RSISOM). The concept of hierarchy self-organizing map, furthermore, is developed to improve the performance of RSISOM for a color image recognition. In the experiment, the proposed algorithm shows the efficient invariant capability under scaling and rotation as well as the distinguish capability in different color-texture images. Furthermore, the computational time after applying the concept of Hierarchy in RSISOM approach is three times less than the computational time of the original RSISOM.