ISSN (Online) 2278-1021 ISSN (Print) 2319 5940 International Journal of Advanced Research in Computer and Communication Engineering Vol. 4, Issue 10, October 2015 Copyright to IJARCCE DOI 10.17148/IJARCCE.2015.41037 184 A Comprehensive Survey of Image Search Based on Visual Similarity Rahul Shroff 1 , Jaykumar Dhage 2 M.Tech Scholar, Dept. of Computer Science & Engg., Maharashtra Institute of Technology (MIT), Aurangabad, India 1 Asst. Professor, Dept. of Computer Science & Engg., Maharashtra Institute of Technology (MIT), Aurangabad, India 2 Abstract: Image search based on visual similarity is the widely applicable image processing method, which is used extensively. One of the important stage in content based image retrieval system is Feature Extraction, where low level features are extracted from image, then The features vector is formed by the extracted features. Feature Extraction is used for indexing images and interpretation of image. Effective storage, ranking and organizing a large image database is a critical issue in computer systems. To overcome these problems many methods has proposed. However, the accuracy and speed of image retrieval is still an interesting topic of research. This paper presents a survey of various states-of-the-art- image search techniques such as color edge detection, Discrete Wavelet Transform and Singular value decomposition etc. that allows faster and effective visual similarity search. Keywords: Feature Extraction, Similarity Matching, Canny Edge Detection, Color Edge Detection, Haar Wavelet Transform, singular value decomposition. I. INTRODUCTION Rapid evolution of multimedia and web technology is the main reason for the growth of images on the internet. Images are the media which is used widely in web pages and hence it created need for wide range of images. There is strong need to retrieve more relevant images from such large image databases. In the past decade content based image retrieval has taken substantial attention over the widely used text based image search engines. A content based image retrieval (CBIR) system has more advantages over traditional image search system, based on image tags, to retrieve images efficiently and effectively. This system helps the user to retrieve relevant images based on visual properties such as color, texture and the shape of an object in the image. In the Place of taking text keyword as input, CBIR systems directly take image query and try to retrieve relevant images from the database based on pre-specified feature space and distance measure. The traditional search methods based on annotations of the images which is very difficult to describe each image with all the possible words, require large amount of space and although not giving satisfactory results. Thus the researches migrated toward CBIR. CBIR does not depend on image annotations or image names. It analyses the content of the image and the search is based on this content. Hence CBIR is a more direct method of image search, giving more satisfactory results, but it is a complex process. The working of the typical CBIR system divided into two major parts. Feature extraction (FE) is the first part, where a set of features are called feature vector, which accurately represent the content of each image in the database. The second part is similarity measurement (SM), where calculate distance between the query image feature vectors and feature vectors of each image in the database and images are ranked based on smaller distance. Effective representation of image features and an efficient search mechanism are two key factors which affect the efficiency of large scale image retrieval system. It is known that the quality of image search results heavily depends on the representation power of image features (vectors).The latter, it is very difficult to develop an efficient search mechanism because existing image features not describe image properly (not give semantic information) and we retrieve similar from huge image database. This paper focus on those techniques which is able to search the very similar images from a huge and possibly distributed image databases. Moreover, search techniques must be memory efficient, able to store billions of images and also do fast similarity search. The remaining paper is organized as follows. System frameworks of Image Search based on visual similarity describe in Section II and Techniques used for similarity image search discussed in Section III. Finally, Section IV concludes this paper. II. SYSTEM FRAMEWORK The flow of image search using its content is shown in Fig. 1. It only needs the user to enter query image with less effort and accordingly retrieve its most relevant images from a given database using smaller distance ranking. The system works in two steps as following: 1) First, visual features are extracted from individual images from database. The features vector is formed by the extracted features. These feature vectors are then stored in feature database. 2) In second, the user has to enter query image for finding relevant images from database. Similarly, it extract features vector of given query image.