Abstract:

A Multi-Points Diverse Density (MPDD) learning algorithm was proposed. In classical diverse density learning algorithm, only one point was used to represent the content of bags, which was not enough. In the proposed MPDD learning algorithm, multi-points were utilized to contain more information. This algorithm was applied to image retrieval. First, an image was divided into small blocks, and the color and texture features of these small blocks were extracted to describe the content of an image block. Then, K-means algorithm was used to cluster image blocks to get instances. Finally, the MPDD learning algorithm was employed to retrieval images. Experiment results show that the performance of the proposed MPDD algorithm is superior to classical diverse density learning algorithm.

Key words: information processing, multi-instance learning, multi-points diverse density(MPDD), diverse density(DD), image retrieval