Multimedia Quality Assessment

One of the problems referred to before is the automatic quality assessment, and in real time, of multimedia streams transmitted through the network. For example, speech quality assessment in IP-telephony applications, video quality assessment in video teleconferencing or video streaming applications, etc. It is a problem since, as we will see in detail in Chapter 3, no previously published satisfactory solution to it exists. Traditionally, the quality is measured as a function of the encoding algorithms, without taking into account the effect of packet networks' parameters. There are two approaches to measure the quality: either by objective methods or by subjective methods. Subjective quality assessment methods [72,70,59] measure the overall perceived quality. They are carried out by human subjects. The most commonly used measure is the Mean Opinion Score (MOS), recommended by the International Telecommunication Union (ITU) [70]. It consists of having a group of subjects viewing and/or listening to processed samples or sequences in order to rate their quality, according to a predefined quality scale. That is, human subjects are trained to ``build'' a mapping between the quality scale and a set of processed multimedia samples. On the other hand, objective methods [154] usually measure the quality by comparing the original and distorted sequences. Some existing methods for video are MSE (Mean Square Error) or PSNR (Peak Signal to Noise Ratio), which measure the quality by performing some kind of simple difference between frames. There are other more complicated measures like the moving picture quality metric (MPQM), and the normalized video fidelity metric (NVFM) [153]. Some examples for objective speech quality assessment are: Signal to Noise Ratio (SNR), Itukura-Saito distortion, Log-likelihood ratio, Segmental SNR and Perceptual Speech Quality Measures (PSQM) [38,59,105]. It must be observed that to verify the accuracy of these measures (also called ``tests'' in this area), they usually have to be correlated with results obtained using subjective quality evaluations.