© 2014, IJARCSMS All Rights Reserved 57 | P age ISSN: 2321-7782 (Online) Volume 2, Issue 5, May 2014 International Journal of Advance Research in Computer Science and Management Studies Research Article / Survey Paper / Case Study Available online at: www.ijarcsms.com Fast Text localization and Retrieval in Image Pradip Shewale 1 Department of Computer Engineering MIT Academy of Engineering Alandi, University of Pune Pune – India Mayura Kinikar 2 Department of Computer Engineering MIT Academy of Engineering Alandi, University of Pune Pune – India Abstract: With the increasing use of digital media, video and images in daily day to day life. Caption in digital media images, capture images or natural images as important for text retrieval in images. In this paper we proposed a fast text localization method in which sobel edge detection and clustering method is used to localize all the possible edges in the image. Optical character recognition method to recognize is the localized edges are text or not. Experimental results show the improvance in the proposed algorithm. Keywords: Caption, edge detection, text localization, text detection, clustering. I. INTRODUCTION With wide use of internet, digital media, digital media capturing devices e.g. mobile phones, cameras people are capturing images, video and upload it to the internet websites like instagram, mypics, youtube etc. Text retrieval in this medium has become research area now a day’s [2]. As it may contain logo, superimposed text, cricket score, information about players, breaking news, Temperature etc. These texts is relevant to the video or image it has many applications like content based web search, logo detection in CCTV video feeds, sign detection, licence plate reading. As visually impaired person can’t see it will useful for them to access text and clustered with text to speech algorithm and make them to read cover of book, labels on door, medicine labels etc. hence caption localization and detection become important research now a days. Our main moto of the proposed system is to detect text in internet video, low quality images downloaded from internet or capturing devices [8]. A lot of research work has been done on text localization and recognition in which method used is grouped in three classes as texture based, connected component based and edge based. Texture based method assumes that text in image have some unique property. Connected component based method assumes that character in image has uniform color while Edge based method assumes that character are made up of edge strokes hence retrieval of edges is important constraint in edge based method e.g. edge detection algorithm[1]. In this paper we proposed a fast text localization method to localize all the possible edges and text in images. Optical character recognition is used to recognize the clustered get from the output of sobel edge detection algorithm [6]. II. PROPOSED METHOD On observation of literature survey and work done on edge detection maps it is observed that text regions contains strong density of edges. As characters are made up of edge strokes they show some inhibit property of finding in pairs and opposite gradient direction. The block diagram of proposed method is as shown in figure. Input to the system is image containing text or .avi uncompressed video file.