Optimal Discrete Wavelet Frames Features for Texture-Based Image Retrieval Applications

摘要

In this paper, experiments were conducted to find the optimal configuration for discrete wavelet frames texture feature extraction method for use in real-time content-based image retrieval application. Several parameters of the algorithm such as the wavelet basis, the number of decomposition levels, and the distance metric are evaluated in terms of retrieval performance, and the optimum value for each parameter is suggested. By experimenting on0020the statistical function as well as channel selection, the final DWF configuration is proposed that achieves an average of more than 80% accuracy using the Bro-datz texture dataset and about 70% accuracy using the VisTex dataset.