International Journal of Engineering and Technology

The development of search engine for similar images stays yet a scientific challenge. Since every image is represented by one or more vectors, the research stage becomes very hard. To raise these difficulties, we suggest inserting a search engine based on multiple representations of images where every image will be represented by three vectors that are color, texture and shape. In the research phase, we applied on the first one clustering on every level of representation for all images stored on the database. This has enabled us to obtain fifteen classes of colors, ten classes of texture and twenty classes of shape. Finally we compute the descriptors associated to the research image and we detect the classes of membership for each level of the representation. Note that the proposed method allowed us to reduce the number of images to check to 7% on average of the database.