JPEG2000 در طول کانال های ارتباطی نویزدار از طریق ارزیابی و تجزیه و تحلیل هزینه

In this paper, we examine the behavior of the JPEG2000 coding scheme over noisy or congested communication channels and highlight a cost policy aspect. Two error schemes are considered, involving bit errors (noisy channel) and packet-dropping (congested channel) effects. Two bit error methods are used, consisting of flipping or eliminating the bits, and various packet sizes are put to the test of packet dropping. Extensive performance results are presented accompanied by an overall cost analysis.

The emerging need for multimedia transmission over high-speed communication systems on which several restrictions and transmission policies apply, guide the research to new ways of data handling and algorithm applications, even before the data can be transmitted. One of the problems in engineering a packet switched network carrying both non-bursty delay-sensitive traffic (voice, video) and highly bursty delay-tolerant traffic (computer data, image data) is the congestion problem [1]. Since digital bitmap representations of images require large numbers of bits, data compression techniques are important for efficient transmission. Standard lossless compression methods, such as the lossless JPEG or the JPEG-LS and JBIG coders, provide with compression ratios of about 2:1 on the average. Unfortunately, such algorithms do not have the ability to allow packet dropping by the network. Hence, when a congested facility drops a packet containing compressed image data, the rest of the image is destroyed, unless the end-user is employing an end-to-end receive—acknowledge transmission—repeat mechanism. Such a protocol saves the transmitted information, but ultimately makes matters worse for the already congested network as it further increases traffic, not to mention the additional disadvantage of increasing transmission delay. Thus, to be effective as a congestion relieving mechanism, packet dropping must be allowed with the knowledge and blessing of the end-user. Presumably such a user would be given pricing advantages for the packets that he/she marks as droppable, since this information is delivered only when the network is idle. Similar problems appear when noisy communication channels carrying delay-sensitive image data change or drop bits. JPEG2000, the new coding standard, comes to fulfill such requirements of progressive coding while providing with error control mechanisms. Several papers and publications consider the performance of this coder in noisy environments in order to compare the scheme with the existing ones [2], [3], [4] and [5]. In this work, we present the results of using the JPEG2000 coder in error resilient mode with Layer–Resolution–Component–Position (LRCP) priority, considering the overall effect of different error models. Preliminary results have been reported in [6]. The outcome of our tests is an overall communication channel cost policy analysis that can be used by providers to impose fees policies and users to evaluate provided services.

In this work, we simulated a communication network assigned with the task to transmit progressive-by-quality JPEG2000 codestreams through a noisy or congested channel. Extensive tests run on the standard test images using various error schemes, modes, rates, burst rates, packet sizes and a variable number of error-free transmitted quality layers gave interesting results concerning not only the error resilience capabilities and restrictions of the JPEG2000 coding scheme, but also highlighted a cost policy analysis aspect of such a communication system. After analyzing this behavior, we defined a “fair” cost network policy and provided with a table of fair cost ratios according to the error rate. Summarizing, we are able to say that: (1) in a noisy channel (bit-flipping environment), • in our error model, corruption of data and quality of decoded image averaged over the burstiness of error occurrence are linearly dependant on the logarithm of the probability of error; • decoded image quality versus data corruption exhibits a parabolic behavior matching a usual rate-distortion curve; • data corruption and decoded image quality versus the number of guaranteed transmitted error-free quality layers, both exhibit straight line behavior with constant slopes (in the case of quality) or slopes dependent upon the error rate (in the case of corruption); • cost of “healthy” data is proportional to error rate, and increases with increasing burst error rates; • cost per quality dB exhibits upper and lower bounds in strict relation with the error rate. Proper selection of cost ratios can result in optimum or “fair” channel operation (from a cost aspect); • error-free quality layers-cost ratio curves, when complementing cost-quality curves, can provide with additional insight aiding in the selection of the appropriate guaranteed error-free bandwidths according to network utilization needs and policies; (2) in a channel with outage periods (bit-dropping environment), • quality of decoded image is lower than the one observed in the bit-flipping mode and shows less dependence on burst error rates; • decoded image quality versus data corruption does not exhibit a parabolic behavior but it rather expresses a linear relation, with quality being almost constant for any value of data corruption percentage; • data corruption and decoded image quality versus the number of guaranteed transmitted lossless quality layers, both exhibit straight line behavior with constant slopes (in the case of quality) or slopes dependent upon the error rate (in the case of corruption), matching the behavior of bit-flipping error mode. Again quality levels are much lower than in bit-flipping error mode; • cost of “healthy” data, cost per quality dB and lossless quality layers-cost ratio curves, exhibit the same characteristics as illustrated in the case of bit-flipping error modes. The differences are the lower quality with a higher cost per quality dB; (3) in a congested channel (packet-dropping environment), • decoded image quality versus data corruption exhibits the same parabolic behavior as in the case of bit-flipping error mode; • decoded image quality versus the size of packets transmitted by the network confirms that in high congestion small packets are preferable; • cost per quality dB versus the error-free transmitted quality layers shows similar characteristics to the ones in the bit-hitting modes. The fact is also highlighted in the error-free transmitted quality layers for minimum cost versus the cost ratio graph.