| With the wide applications of informational technology and all kinds of digital medical imaging equipments in the medical field, a large number of digital images are produced in hospital. How to manage and use the images efficiently is a hot topic. So the content-based image retrieval (CBIR) is appeared, which is different from the text retrieval, and the application of CBIR to the field of medical images have great significance.The main challenges in CBIR are due to the two gaps. The first is the sensor gap between the object of the world and the information represented by computer, and it was solved by feature extract. The second one is the semantic gap between the low-level visual features and high-level human perception and interpretation. Relevance feedback technique was employing to bridge the semantic gap.In the relevance feedback, reseachers reference the idea of re-weighting which was firstly used in text retrieval. Then the machine learning techniques were introduced in CBIR, for example, SVM. With the elapse of time, long-term learning was used to record the history informationof retrieval for getting relative information.In this paper, we first introduce the key techniques of the CBIR, especially analyze the current relevant feedback algorithms in detail, and then present a new relevant feedback method based on hybrid methods which combines the long-term learning and short-term learning methods. First, the short-term learning is based on Re-weighting and Soft SVM. With the analysis about all kinds of selecting, we present a general algrothim for realize the different selecting by regulate the parameter. Next, for the long-term learning we use the log-based method, where the multi-level-labeled information and the image ID was recorded in the log-database. Extensive experiments are designed and conducted to evaluate the proposed algorithms. The results show that the hybird method is effective for CBIR of medical image. |
PDF Full Text Request