OBJECT DETECTION AND LOCALIZATION IN THE WAVELET DOMAIN

Abstract

We are interested in the problem of detecting and localizing objects in the compressed domain. The practical uses of this research are video surveillance, queries over digital library archives and teleconferencing. Most image operations, such as object recognition, are formulated as sequences of operations in the image domain. Such methods need direct access to pixel information as a starting point, but pixel information is not directly available in a compressed image stream. The standards that have emerged for still-image and video compression each contain steps that are commonly found in compression algorithms, like linear transformations, coefficient quantization, run-length coding and entropy coding. Coders like JPEG 2000 and SPHIT are built around the wavelet transform. Thus as a step toward detection and localization of objects embedded in the compressed bit stream we consider here the problem of localizing and detection in the wavelet domain.