Two-layer image retrieval method based on wavelet and local color spatial features

An improved method based on two layers image retrieval is proposed. The first layer retrieval extract edge of wavelet modulus maxima and use 7 moment invariants as the shape features and the averages and variances of the multi-scaling 2-D wavelet decomposition coefficients as the synthesis eigenvector. The second one consists of the local color histograms. Difference weights are assigned to every block according to the importance of its spatial distribution. This method combines the advantages of the 7 moment invariants and the local spatial features, thereby achieving the better retrieval performance.