Content-based Image Retrieval (CBIR) using relevance feedback technique is applied to improve the results of traditional techniques in image retrieval. Since the results returned by system cannot fully satisfy users and the iteration process of feedback can be very time-consuming and tedious, log-based relevance feedback is introduce to the system. In previous work, we have already introduced multi-level log-based relevance feedback scheme for image retrieval to accelerate the iteration process and to increase the hit rate. In this paper, we improve the novel algorithm and apply it in a demo image retrieval system which presents refined results based on multi-level log-based relevance feedback for Content-based Image Retrieval.