Brushlet Features for Texture Image Retrieval

In this paper, we employ a new adaptive basis of functions — brushlets for extracting texture properties. Brushlets are functions which are well localized with only one peak in the frequency domain. Hence, a representation of texture in terms of spatial frequency distributions can be constructed. The Brushlet features are used in texture image retrieval experiments to assess its effectiveness by comparing with retrieval results obtained using other commonly used wavelet and Gabor based representations. Experiments using the Brodatz texture database indicates that the brushlet features achieve a very good retrieval comparable to and slightly better than that using Gabor features and better than the wavelet features. The advantage of the brushlet features is that they require far less computation to extract than the Gabor features for achieving comparable performance.