Effective content based image retrieval using various feature extraction strategies

Abstract

Searching of digital information on the Internet has become an important part of our daily life There is an exponential increase of visual content available on the Internet and there is a need for systems that are able to search images and quickly display the relevant images Image retrieval systems have been developed to address this need of the databases which is a new challenge that one has to face COREL10K image database is considered and the Mat lab tool box is used as a simulation tool In our first framework Cubic Spline Neural Network CSNN architecture is employed to determine the non-linear relationship between image features so that more accurate similarity comparison between images can be supported Eight feature values are extracted from a given query image namely three colour components Hue Saturation Value three colour moments Mean Standard Deviation and Skewness colour corelogram and texture feature The texture feature is one among the important features in content based image retrieval and is obtained using wavelet transformation The Daubechies4 wavelet filter is used for Discrete Wavelet Transformation of images All these features are extracted from the database and query image Similarity measurements are carried out to retrieve the relevant images for the given image query The accuracy of this method has been found to be fairly good In the second framework the quad tree block truncation coding is used for compression and features are extracted from the encoded images The novelty of the proposed work is based on the concept of quad tree decomposition That is the given image is divided into four blocks and again subdivided into four blocks until it reaches the minimum difference by using newline