Fine-grained correlation analysis for medical image retrieval

Abstract Feature fusion in medical image retrieval remains a challenging task because of the high-dimensional data and massive amount of irrelevant information in images. To solve these issues, we propose a novel feature fusion method, called fine-grained correlation analysis (FGCA), for medical image retrieval. First, we analyze the problem that there are many irrelevant local regions in a category. To solve this problem, an image is partitioned into some fine-grained samples. Then, the fine-grained samples with similar characteristics are tagged with the same label by the k-means clustering algorithm. Finally, we investigate how the correlation relationship extracted from the fine-grained samples helps fuse different features and obtain the more discriminative and less redundant information for medical image retrieval. Experiments on three medical image datasets show that our proposed FGCA approach works better than the conventional methods.