New Texture Features Based on Wavelet Transform Coinciding with Human Visual Perception

Wavelet decomposition is widely used in texture image retrieval and classification. Subbands statistics are normally used as feature vectors during feature selection phase. However, most previous methods seldom make further analysis of the transforms' subbands according to human visual perception. Based on wavelet decomposition, this paper proposes three new texture features which coincide with human visual perception. These features include directionality, contrast and coarseness in wavelet domain. We test our new texture features using the Brodatz texture database, and the experimental results indicate that the new texture features are consistent with human visual perception well.