Resampling HD Images with the Effects of Blur and Edges for Future Musical Collaboration

Image down/upsampling can give significant bit rate reduction with hardly noticeable quality degradation. This is attractive to our vision of realtime multiparty collaboration from distributed places with video data at HD resolution that must guarantee maximum end-to-end delay of 11.5ms to enable musical synchronization. Based on the Distributed Multimedia Plays architecture, this paper compares the performances of bicubic and Lanczos techniques as well as one recently proposed for image upsampling in terms of computing time, level of blur, and objective image quality in PSNR and SSIM. The effects of image blur and edges to resampling are also examined. The results show that the classic Lanczos-2 and bicubic techniques perform the best and thus are suitable for the vision due to their potential for efficient parallel processing in hardware. We also show that composite images with different downsampling factors can achieve 4dB increase in PSNR.