Subband coded image reconstruction for lossy packet networks

Packet-based transmission of subband coded images over lossy networks presents a reconstruction problem at the decoder. The paper presents two techniques for reconstruction of lost subband coefficients. The low frequency reconstruction algorithm maintains smoothness and exploits intraband correlation by fitting a surface to known coefficients to interpolate lost coefficients. Accurate edge placement is achieved by warping the interpolation grid based on local high frequency characteristics. High frequency reconstruction is performed using linear interpolation, providing good visual performance and maintaining properties required for edge placement. The algorithm is applicable to any number of decompositions, both luminance and chrominance components, and can be used with progressive transmission. Computational overhead is minimal at 0.5% per percentage of coefficients lost and the reconstructed synthesized images maintain good visual quality at loss rates as high as 10%