IMAGE QUERY BASED SEARCH ENGINE USING CONTENT AWARE IMAGE RETRIEVAL

Successful Content-Based Image Retrieval is a challenge even today. The present system for image retrieval is quite inefficient and expensive. This paper proposes a system that deals with searching and localizing all occurrences of an image query. The image is constituted using a set of descriptors which makes recognition successful even when there is a slight difference in illumination or a change in point of view. A dictionary of visual words is created through raw descriptor matching by trying to utilize the same concept as text retrieval approach from a document. This content-aware Image retrieval system forms full frame queries from a region, matches it with the image query. This process uses the Scale Invariant Feature Transform (SIFT), a feature detection algorithm to outline features in an image. The query image is recognized in the database by individually comparing each feature and grouping the similar images by calculating Euclidean distance of their respective feature vectors. This process is trained on a pre-determined (nearly 24,000) number of image frames of a group of episodes of the F.R.I.E.N.D.S T.V. show by extracting 24 frames per second and dropping the common frames