Semantic Memory Learning in Image Retrieval Using k Means Approach

Abstract

To reduce the conceptual gap in content-based image retrieval (CBIR) and small training problem in relevance feedback (RF), this paper attempts to focus on the semantic memory learning in image retrieval using proposed 2-means clustering. In this system, initial retrieval results of CBIR are obtained, and then user’s opinion is given to the system as relevant/irrelevant to the user. With this user feedback, we can easily make the relevant image cluster and the irrelevant image cluster directly instead of random selection. Hence with initial known clusters and number of clusters, computational time is highly reduced for finding cluster center. We have also reduced the burden of clustering by considering only relevant cluster repeatedly for each feedback iteration. We experimented on two different data sets using proposed system. Results are found better compared to the earlier approaches.