Article Preview
Top1. Introduction
The field of information technology has known significant innovations since few decades. With the emergence of web development and transmission networks, the amount of images available to users continues to grow. The result is a permanent and considerable digital image production in many areas such as architecture, satellite imaging, video surveillance, robotics, medicine and health, illustration, audiovisual, botany, etc. this is due to the appearance of the image acquisition devices that produce each day a very large number of images. Thus, this mass of data would have no interest if we could not easily find the information concerning a particularly interest. This sparked a need for development of multimedia information search techniques, especially in image search. The list of possible applications of image search by content is immense. While it is important to model images, compress them, store them and transmit them, it is also important to develop efficient systems for handling, classifying and indexing these images and access to them quickly in the databases of images.
Images indexing and searching is a computer research field that is booming promoter, and a topical theme. It aims to define models and systems whose purpose is to offer the possibility to the users to access, manipulate and exploit directly the gigantic image databases using only their content. This explains the intensive research activity devoted to this field. Thus, the field of image indexing and searching is centered on users, especially through the notion of relevance.
The need for indexing and searching methods based directly on the content of the image is not so well known. Among the developed systems, we find the approach based on the manual annotation text of images known as “Text Based Image Retrieval (TBIR)”, and this method uses a database management system to manage these images using a query language such as SQL. Through textual descriptors, images can be organized hierarchically according to the semantic and themes to facilitate search and navigation in the database.
Despite the great success of this approach to research documents, textual annotation of images showed its inability to satisfy the need for information covering the actual content of the images, and it is a long task, heavy, expensive and repetitive for the user.
To overcome the drawbacks of images search systems based on the manual-textual annotation, image search system based mainly on the content that aimed extracting relevant information directly from the image is proposed as an alternative solution to the textual approach. This approach began in the early 90s and it is nowadays the subject of intensive research. it is often called “Content Based Image Retrieval” (CBIR).
Recently, the research focuses on CBIR systems that is fetching the exact cluster of relevant images and reducing the elapsed time of the system. For this purpose, various techniques have been developed to improve the performance of CBIR system. Clustering is one of them. Clustering is the technique which is used to partition the data into groups of similar objects. Note that CBIR systems use usually three low-level primitives to extract inherent information based on color, texture and shape descriptor.
The intent of the classification process is to categorize all objects in a data base color image into one of several classes, or “clusters”. This categorized data is based to their similar features present in a database image. Image classification is a labeling process, in which image pixels are categorized into different classes in case of single image, but in case of database image instead of pixels we take an image as an object and categorized into different classes. Note that image classification algorithms have been successively applied to a range of problems, including image segmentation and color quantization, change detection for land cover monitoring, and data mining among others.
The proposed system uses classification based on k-means clustering algorithms which are perfectly suited to the main difficulty residing in the optimal features selection that are able to produce clusters with spatial homogeneity.
Our main goal in this work is to set up an algorithm that can retrieve similar images of a given image (query). Toward this end, we have developed and tested our algorithm for image retrieval using two image databases. Here we use Euclidean distance to match images and to measure accurate similarity for image retrieval. We have successfully applied our proposed method to image retrieval using only visual image content.
The rest of the paper is organized as follows: section II deals with related works. In section III, we develop the mean steps of the proposed method. Section IV presents the experimental result. Finally, section V concludes the paper.