Streaming video with optimized reconstruction-based DCT

One fundamental problem with streaming video data over unreliable IP networks is that packets may be dropped or arrive too late for real-time playback. Traditional error-control schemes are not attractive because they either add redundant information that may worsen network traffic, or rely solely on decoders with inadequate error concealment. This paper presents a joint sender-receiver approach in designing transforms for multiple-description coding in order to conceal network losses in streaming real-time video over the Internet. On the receiver side, we adopt a simple interpolation-based reconstruction, as sophisticated concealment techniques cannot be employed in software-based real-time playback. On the sender side, we design an optimized reconstruction-based discrete cosine transform (ORB-DCT) with the objective of minimizing the mean squared error, assuming that some of the descriptions are lost and that the missing information is reconstructed by simple averaging at the destination. Experimental results show that our proposed ORB-DCT performs better than the original DCT in real Internet tests. Future research includes finding a perceptual-based quantization matrix based on extended basis images derived for reconstruction, and incorporating the effects of quantization and inverse quantization in the design