| Summary: | In this study, we propose a novel dictionary learning based multi-level clustering method for content based medical image retrieval. We suggest innovation compared to previous works as the center of the information-part of the image is calculated, the entropy of pixel intensity values is used as feature in the regions of the image and the K-SVD method is used to generate dictionaries for each cluster. Then, the process is repeated in each cluster in order to form sub-clusters. The performance of the proposed method is evaluated using the ImageCLEF Medical Dataset. Our feature extraction methods use image region partitionings that aim either at providing rotation and translation invariant CBIR, or at taking into consideration the rich information usually available at the center of each medical images. Afterwards, we compare the performance among the different feature extraction methods as well as the pure k-means algorithm and present our findings.
|
|---|
