Successive Image Transmission

Successive image transmission is a technique which enables an approximate image reconstruction quickly, with fine details to be transmitted successively by several passes over the image. This technique is particularly useful for image transmission over low bandwidth channels and browsing through remotely stored images. An important aspect of a coding method for successive image transmission applications is the efficiency with which useful contextual information in an image can be transmitted. Segmentation-based approaches are promising in this context because much of the semantic information in the image is in contours, which are an integral part of the code. Therefore, segmentation-based approximation allows very efficient transmission of semantic information, which can help the recipient in deciding quickly whether to abort the transmission of an undesired image or to continue with successive transmission.

Using our segmentation-based approximation for successive image transmission, the user may first be presented with a contour map. On reception of the contour map, the user can quickly decide whether to request the next level of approximation (regions filled with mean values) or to abort the transmission. Since, the number of bits required encoding of the contours is small, it can be transmitted very efficiently. As a next step, the mean intensities which fill the regions in the approximation can be retrieved. This can be accomplished very efficiently by transmitting the intensity list and filling the regions in the contour map with the corresponding mean intensity value. If a higher visual quality reconstruction is desired, the quantized error can be transmitted progressively as bit-planes and added to the approximation. In this fashion, the user can view the image from a coarse approximation to a reconstruction of perfect visual quality. If numerically perfect reconstruction is required, this can also be achieved (with appropriate quantization of the error).