Image transmission over error-prone channels: sigma filtering andmultiple description objectives

摘要

This work considers a fundamental problem of image transmissionover error-prone channels. The impairments that we target are long bursterrors and transient channel failures, as would occur in the wirelessnetwork because of an obstruction in the transmission path, or severetraffic congestion in a packet network such as the Internet. Theseimpairments cause many packets to be dropped. In contrast to thetraditional schemes that first introduce some amount of redundancy amongdescriptions, and then use this correlation for combating channelimpairments, we investigate a new scheme to tackle this problem. Theproposed scheme is to capture the most important visual features of agiven image, extracted by sigma filtering preprocessing, intoindependent equal length packets. At the receiver side, we adopt a smartinterpolation-based technique. The reconstruction quality at thereceiver depends only on the number of packets received, but isindependent of the place from where they were cropped. The preliminaryresults on standard test images show that our reconstructed image isvery pleasant to human eyes, and achieves reasonable PSNR values, whileat the same time using a lower bit rate than other reported traditionalschemes based on multiple description coding