Abstract: Description of content as well as semantics is important in Content Based Image Retrieval. Even though the image gets interpreted semantically, the retrieval accuracy of CBIR systems is often low. Many CBIR systems still rely on text retrieval technologies on human labeled keywords. Identifying and Learning the interest point of humans are critical components in Image reclamation. This manuscript advises a Neural Network scaffold that incorporates unsupervised learning in to query refinement process. In order to indentify the user interest point we have used this as a relevance feedback approach which easily maps stumpy characteristics with user’s lofty concept. Experimental results illustrate the effectiveness of this approach.