Internet image indexing and searching

This paper presents the framework of our WWW image indexing and searching system. The semantics of an image are obtained by analyzing its associated texts. During the processing, stopword-removing and term-stemming algorithms are applied in order to reduce the noises and highlight the semantic concepts of images. In our solution, the semantic weight of a term is calculated with respect to its tag type, frequency of occurrence, and its relative position to an image. The weight of a term will be increased if other terms close-by have some semantic relationships with it, such as SYN (synonym), PART-OF, or IS-A. Color histogram, texture, and the shape are selected as the visual features and extracted in order to support visual-oriented retrieval.