Joint source channel distortion modeling for JPEG images

We have developed a distortion model for predicting the amount of distortion in progressively compressed Joint Photographic Experts Group (JPEG) images due to quantization and bit errors when transmitted over noisy/fading channels. The images are divided into different quality layers, with each layer containing DCT coefficients from a particular subband. Distortion expressions are derived for the DC layer and the AC layers separately, with the total distortion being the sum of the distortions due to individual layers. These distortion expressions represent distortion over a set of images rather than an individual image. Important source coding components such as differential coding, run-length coding and entropy coding are also taken into account by our model. Model parameters are derived from a training database of images. These parameters are then used to predict the distortion in a test database of images at various source coding rates and channel bit error rates. The model results are compared to that of actual simulation results. These results show that this model predicts distortion to within 1.5 dB PSNR at all the source coding rates and channel bit error rates.