Quality of video affected by packet loss distortion compared to the predictions of a spatio-temporal model

Image quality of images as judged by human observers may be determined by different simulation models. The results of an image discrimination model is discussed and compared to the results of the peak signal-to-noise ratio. The image discrimination model simulates how people analyze spatial-temporal information and it predicts detection of distortion. Peak signal-to-noise ratio measures the physical difference between an original and a distorted image sequence. To test if the model could without modifications be used to predict image quality in small moving images, a perceptual experiment was carried out with ten observers. Image quality judgments were measured with five different video scenes using eight category scales and magnitude estimations. Packet loss of transmissions was simulated for two H.263 coders, one with two layers and one with only one layer. The reliability of the judgments was generally high. The judged image quality depended on type of scene and coder. There was a strong inter-correlation between the category scales. Both magnitude estimations of image quality and ratings of a category scale for image quality could to some extent be predicted by the model, but there were no advantages for the visual model.