Enhancement of content Based Image Retrieval with Combined Colour Straight Line and Outline Sketch Signatures of the Images

Abstract

newline viii newlineABSTRACT newlineThe primary aim of this research study is to enhance the performance of Content newlineBased Image Retrieval (CBIR) systems. An elaborate survey on existing CBIR newlinesystems and related technologies has been carried out. newlineLarge number of images are created and stored everyday due to the advances in image newlineacquisition technologies and data storage techniques. In order to handle these data, it newlineis necessary to develop a suitable information system to manage efficiently. Image newlinesearching and retrieval of desired images is one of the most important services that newlinesupport a system with image collections. Some of them are educational image newlineresources, medical images, fingerprints, surveillance and security systems, satellite newlineimages, photo collections, museum pictures etc. Content Based Image Retrieval newline(CBIR) system searches the most similar images as that of a query image from the newlinedatabase by comparing the feature vectors of all the images in the database with that newlineof the query image. newlineThe four processes that take place in retrieval of desired images from the database are newlineImage Pre-Processing, Signature Extraction Processing, Query Image Processing, newlineImage Ranking based on the similarity between the query image and the database newlineimages for Image retrieval processing. newlineIn image preprocessing, colour features are extracted from top, middle and bottom newlineregions of image and represented as histograms. All the three histograms are newlinecombined to form the histogram H which indirectly gives spatial data categorization. newlineColour, Straight line, Outline sketch and Texture signatures are extracted in signature newlineextraction processing. All these signatures are combined, indexed and stored along newlinewith the images stored in the image database. The extracted features are trained with newlineSVM neural network and classified into different categories. Thus supervised machine newlinelearning method is used to categorise the images. In the query image processing, newlinequery for retrieving relevant images from the image database are given as image in newlinetwo ways. It may be image or any one image from video by converting video into newlineframe. Any one frame is given as query. The given query image is categorized by newlineix newlineSVM neural network and those images in the database which have the same category newlineas that of query image are considered for similarity measure. The similarity between newlinethem is measured calculating the Euclidean distance between their features. newlineAccording to the similarity measure, the images are ranked. Images are sorted in the newlineascending order based on ranking. Top N ranked images from the image database newlinecorresponding to the index are retrieved as the similar images. newlineEight methods were developed for the image retrieval such as newlineand#61623; CBIR using 3 Region Colour Signature (3RCS) method newlineand#61623; CBIR with combined Colour and Straight Line Signature (CSLS) method newlineand#61623; CBIR with combined Colour, Straight Line and Outline Sketch Signature newline(CSLOS) method newlineand#61623; CBIR with combined Colour, Straight Line and Outline Signature using newlineMahalanobis distance method (CSLOSM) newlineand#61623; CBIR with combined Texture, Colour, Straight Line and Outline Signature newline(TCSLOS) newlineand#61623; CBIR using Colour Co-Occurrence Matrix with combined Colour, Straight newlineLine and Outline Signature (CCMCSLOS) method newlineand#61623; Improved CBIR with combined Colour, Straight Line and Outline Sketch newlineSignature (ICSLOS) method by refining Euclidean distance measure newlineand#61623; SVM based CBIR with combined Colour, Straight Line and Outline Signature newline(SVMCSLOS) method using Euclidean distance measure. newlineThese CBIR methods were tested with benchmark data base called Semantic newlineIntegrated Matching for Picture Libraries (SIMPLIcity) using MATLAB software. newlineSVM based CBIR with Combined CSLOS method using Euclidean distance measure newlineis the proposed method which provided better performance of retrieval of images newlinefrom the database. newlineThis method was also experimented with the video of moving aircraft. Successful newlineimplementation of this CBIR method could be an effective tool for any surveillance newlinesystems that could contribute to the aerospace and defence sectors. Similarly the newlineproposed SVMCSLOS method was experimented with the video of moving bus newlinex newlinewhich can be used in transport applications. The approach of the present study had newlineconsiderably enhanced the performances of CBIR of images in various applications newlinecompared to hitherto methods followed, thus taking this research to its logical newlineconclusion of applicability. newlineAt the end of the study, SVM based CBIR with CSLOS method had emerged as the newlinebest option for retrieval of images; considering various performance parameters. newline