Feature Extraction from Rural Satellite Imagery Using Color Based CBIR Techniques

Abstract

The ability to organize and retrieve visual information such as images is becoming a crucial problem for specialists, because processing visual information requires perceptual abilities not yet known to exist in computational form. Therefore the ability to retrieve visual information is a complex problem. The main retrieval pattern include the features of an image such as colour, and shape. The algorithms used to calculate the similarity between extracted feature with an image database, that utilizes colour features of the images as the basis of comparison and retrieval. An object-oriented definition also involves prescribing a set of pertinent features or pixels as well as a method such as a classification algorithm with the appropriate training data. A low resolution satellite image of a rural scenario has been selected for the analysis. The entire image scene is divided into four equal quadrants. The Content-based retrieval of image features have been carried out for the selected two quadrants I and II. The color classification of the image in two quadrants clearly indicates that there are four important features that are common to the images 1 The features identified include, Natural Vegetation, Water Bodies, Land, and Housing associated with the image. Further using the L*a*b color spectral distribution and histogram techniques an estimation of the different features identified in the image has been made for all the four quadrants. The results are presented in the paper.