JPEG-XS error resilience enhancement

The ever increasing demand for Low Latency Ultra High Definition video streaming applications raises major challenges for the Video over IP industry. In particular, this type of applications requires very high data rates. Although compression is able to decrease these data rates, it generally induces a significant latency which is not compatible with Real-Time applications. However, JPEG-XS has recently been introduced in order to solve this issue as it has been designed to provide a lossless quality at a fair compression ratio and at a very low latency. For this reason, JPEG-XS has become the codec of reference for real-time video streaming. Recently, applications targeting the wireless transport of UHD frames, such as wireless TV or wireless Virtual Reality, have received a strong interest. Those applications raise some challenges due to the unreliability of wireless networks as they suffer from packet loss. As JPEG-XS error resilience against packet loss has never been investigated, this thesis evaluates it and describes some techniques that could be used to improve it. More precisely, two complementary solutions are presented. The first one is an error resilient implementation of the JPEG-XS decoder which is designed to overcome the limitations imposed by the JPEG-XS standard itself. Conceptually, this new decoder tries to limit the frame corruption to the codestream elements which have actually been corrupted instead of propagating it through the entire codestream. In this thesis, we show that this new decoder is able to output frames of a better visual quality. Indeed, at 10% of packet loss rate, the average PSNR is increased by 13dB. The second solution consists in an Unequal Error Protection scheme based on shortened Reed-Solomon codes over GF(256). By a careful analysis of the JPEG-XS codestream content, we have been able to isolate the elements which have the most significant impact on the reconstruction of the compressed image. By applying different amount of redundancy to the different parts of the codestream and by using some specific features of JPEG-XS, we have been able to further increase the visual quality of JPEG-XS encoded images on lossy networks. Actually, at a packet loss rate of 10%, our Unequal Error Protection scheme is able to further improve by 6dB the average PSNR.