Progressive image transmission by segmentation-based coding.

Abstract

Progressive image transmission, where an image builds up gradually, is gaining popularity in image database browsing applications. In these applications, a user might have to reject a large number of unwanted images before selecting the desired one. In such a case, the time required to identify the image contents becomes very important. In this thesis, we present a novel progressive image transmission technique based on a representation by segmentation. The technique preserves edges at low bit rates and is biased toward fast identification. An image is segmented into regions having constant intensity by applying a morphological operator: the watershed. The segmented image is gradually simplified using a graph. The simplifications are transmitted in the reverse order. At the decoder, the image is dynamically divided into an increasing number of regions as the transmission progresses. A subjective experiment was designed, and the recognition times of images transmitted with the proposed algorithm were compared with the recognition times of the same images transmitted with JPEG. The proposed method was found to result in faster recognition of image contents for almost all the images. Some work on the objective evaluation of coarse images is also presented.