Application of Relevance Feedback Based on Rocchio Theory for Medical Image Retrieval

摘要

The preliminary research indicates that color auto-correlogram based on FCM (Fuzzy C-Means) clustering displays good effect in endoscope image retrieval and semantic feature can reduce the "semantic gap." Herein, a feedback method based on Rocchio theory is further discussed. First, a multi-level feature fusion algorithm including low-level color feature and high-level semantic is proposed. Then, Rocchio feedback method is combined with the feature fusion algorithm in order to enhance retrieval efficiency. Finally, the experiments on an image database composed of 1361 endoscope images give encouraging results.