Robust affine invariant shape image retrieval using the ICA Zernike Moment Shape Descriptor

In this paper, we proposed a new affine invariant region-based shape descriptor, the ICA Zernike Moment Shape Descriptor (ICAZMSD). IndependentComponent Analysis (ICA) is first used to turn the original shape into a canonical form, in which the effects of scaling and skewing are eliminated. Next, the properties of the Zernike transform is used to further eliminate the effects of any possible rotation and reflection of the canonical shapes, in extracting the Zernike moments as the affine invariant region-based descriptors. Using the proposed ICAZMSD as shape feature, shape-based image retrieval experiments on a 4000 complex shape image database and a 5600 simple shape image database, show promising retrieval rates of 99.80% and 92.25%, respectively.