Content Based Image Retrieve: Relevance Feedback

Abstract

Increase the use of images and video in an entertainment, education, commercial purpose, research etc. So increase database size of images and video. It’s necessary to need abstraction for efficient and effective browsing of images. There are various models are used for image retrieve. Content Based Image Retrieve is one model for retrieving of images based on low level features as color, texture, shape. Color is inevitable feature that dominates the human perception most. Global color dominance and local color information are used for color features. Quantizing images in color and pixel space to optimize in memory need and performance. Texture information is obtained using co-occurrence matrix of the color images converted to gray. The properties like correlation, energy, homogeneity and contrast are good parameters to measure textural property; these are evaluated at proper offset and angle. For different users have different results in their minds. The semantic gap is generated between low level features and high level perception. This problem is solved by Relevance Feedback. After retrieve images based on low level features, user select most relevant images and give feedback of relevant images. Update the weight of images based on user’s feedback. Two different methods are used for relevance feedback. One method of relevance feedback is simple apply low level features of given feedback images. Another method is Support Vector Machine method which is used the risk minimization and binary classification principal.