Scene retrieval with color moment invariant

In this paper, a novel scene retrieval method has been proposed. Under this framework, homogeneous color regions are detected and then described by color moment invariants. Different from most existing methods, color and spatial information are combined as the uniform descriptor to capture the image property. Moreover, the image descriptor in our method has alterable data structure, whose size is adaptive to the context of the image itself by a two-stage clustering technique. Experimental results show that this method has improved efficiency and robustness of color based image retrieval methods, without the cost of simplicity and compactness.