Image retrieval and characterisation by spatio-chromatic information

Abstract

The wide spread use of images attracts extensive studies in image indexing and retrieval in recent years. With dynamic growth of digital communication, the retrieval of image from thousands to millions of database is becoming a difficulty. There is an urgent need for an effective image-similarity algorithm for image retrieval. Image-similarity measurement plays a key role in developing image-retrieval system while human visual sensitivity provides the essential elements to construct such similarity algorithm. In order to have a better understanding of human perception in image-similarity judgment, numerous experiments were conducted for evaluating the similarity measurements of various complex images. This thesis proposes a hybrid image-retrieval model, the Spatio-chromatic Map, using the achromatic orientation-scale map to represent the spatial information and the 2D chromatic colour histogram intersection to represent the colour information. With an effective image feature representation, the spatio-chromatic map works successfully for retrieving image and predicting human perception in image-similarity measurement, regardless of image databases.