Contents I. Introduction
In this current digital world, the volume of visual multimedia data like image and video are increasing rapidly day by day due to the advancement of several image capturing devices like camera, smartphones etc. These visual multimedia data become an essential part of various applications like medical, education, engineering, architecture, graphics and much more. The typical size of an image is comparatively huge than the textual data to the proper storage and same time efficient retrieval of this large volume of data from any large database is a tedious task.. In this paper, we have addressed some issues related to the image retrieval process. In an image retrieval process, the users are looking for some relevant images from large digital repository within adequate time. The straightforward image to image searching is not a suitable approach in image retrieval due to their large dimension. In literature, numerous of text based image retrieval (TBIR) [1] processes are suggested. In TBIR, each image in a database is attached with a descriptor text or strings like image name, image sequence etc. Now, images, which match most closely to the attached string, are the retrieved output but this annotation process is carried out manually or automatically so it does not give proper description of content of the image. However, in content based image retrieval (CBIR) [2], every image in the database is represented by feature vector where the feature vector consists of salient visual components of an image. Even the size of this feature vector is very small compared to the input image size. So, this feature vector is more suitable in image retrieval process as its dimension is small as well as it is identified by visual features. Naturally, this feature vector provides more relevant information without human involvement for visual perception in the annotation process.
