Implementation of vector based image annotation for content based image retrieval

As the database have huge amount of images and other files it is difficult to retrieve the required image. The performance of existing method is not up to the mark and accuracy is also low. The novel approach identifies contents of the image using provided tags and each tag is associated with a class. Annotation is made based on the vector values. Here whole image is considered as one segment and then extraction process is employed. Currently available datasets contain a small number of classes such as cars, buildings etc. Hence LabelMe database is used to annotate the image. It provides functionality such as drawing polygons, querying images and browsing the database. Using this tool user is free to label as many objects depicted in the image as they choose. The resulting label is stored in the XML file format, which makes the annotations portable and easy to extend. The labeling is to be made only once while storing the image in the database. Set of Precision and Recall values are used to evaluating the performance of CBIR system. User gives the query as input to the system which is actually the content in the image. To retrieve the image it tag each segment with other, and search image in database, if match found then required image will be given as output. As images take more time and space while transmission, this system sends tag file that is in XML through network to reduce time and space during transmission.