Performance evaluation of DCT, Walsh, Haar and Hartley transforms on whole images and partial coefficients in Image Classification

In recent years image processing tools and techniques are gaining continuous attention and recognition due to augmented usage of still and motion images in vivid fields like medical science, engineering, mass-media and GIS. Special category of image processing technique called “Image Classification” has got pivotal position in all the above said fields. An image classifier is a system which accepts query image as an input in part or whole form associates it with one of the stored classes and returns its class label/number as an output. Image classification helps in managing large amount of image data and easy retrieval of image(s) of interest using some kind of similarity measures like Euclidean Distance(Ed), Mean Square Error (MSE) etc. This paper presents broadly used image classification technique using the concept of feature vectors of partial coefficients using Discrete Cosine, Walsh and Haar and Hartley transforms. Partial coefficients of sizes 8×8, 16×16, 32×32 and 64×64 are used. Euclidian distance and MSE are used for similarity measures. The system uses two databases of 500 images each spread over 10 different categories. The results of various transforms on original and the portions of images are compared.