An image retrieval system based on identifying objects of interest

Abstract

This paper presents an effective content-based retrieval mechanism based on previous work in the area of image processing and segmentation. We use the Haar transform to isolate the object of interest (OOI) in the image and then remove the effect of rotation using the Hotelling transform. Next we find the square that best fits the OOI and rescale it to a normalized size. Finally, the 2-layer 2D wavelet transform is computed, and the approximation coefficients are saved as the feature vector to be used at a later stage for retrieving images. Experiments show that the system is invariant to different image properties. Also the retrieval efficiency of the system has proved better than other retrieval systems.