IMPROVING PERFORMANCE OF CONTENT BASED IMAGE RETRIEVAL SYSTEM USING FEATURE PROCESSI
project report helper|
Active In SP
Joined: Sep 2010
22-10-2010, 02:55 PM
inistial pages.doc (Size: 46 KB / Downloads: 71)
IMPROVING PERFORMANCE OF CONTENT BASED IMAGE RETRIEVAL SYSTEM USING FEATURE PROCESSING AND RELEVANCE FEEDBACK
Content based image retrieval (CBIR) system is responsible for searching image database for specific image features like, color, texture and shape. Feature Processing includes extracting different features from the images and then forming a weight vector from the extracted features by normalizing these features. Relevance feedback is the repetitive process of refining the query, which is enhanced on the basis of previously retrieved results by the user.
After taking insight in to different techniques implemented in the simple context of CBIR system and relevance feedback architecture, we are able to visualize a new approach that allows user to completely explore the image database, even when user is not holding relevant image.
The technique is composed of three different phases. Initial phase is about clustering the images in to different groups, Second phase is responsible for selecting features of different images and fine tuning the query through using different types of visual features. Third phase comprises of automatic relevance feedback which works well throughout several steps.
We have built an initial CBIR system, on which we can perform our experiments and have implemented the third phase of the proposed technique by comparing different techniques. We have also compared the performance of two different features color and texture and have found that precision of texture query gives better retrieval results as compared to color query. We have also compared three different relevance feedback techniques using texture feature and found that technique of Multimedia Analysis and Retrieval System (MARS) is giving better results.