Moving Pictures Quality Metric (MPQM)

MPQM is based on a basic vision model. The input to the metric is an original video sequence and a distorted version of it. The distortion is first computed as the difference between the original and the distorted sequences. The original and the error sequences are then decomposed into perceptual channels by the linear transform stage. It uses a bank of Gabor filters [135]. There are seventeen spatial filters and two temporal filters. From this stage, a total of 34 perceptual components for each sequence are to be computed. The next stage models pattern sensitivity, using a contrast sensitivity function estimation and a linear summation model of masking. They are to be computed for every pixel of each perceptual component of the error sequence. The contrast threshold values are then used to divide the perceptual component of the error sequence (on a pixel-by-pixel basis). The data is then gathered together to yield a single figure and to account for higher levels of perception, which is termed pooling. This step is computed as follows. First, the pooling is computed over blocks of the sequence. Such blocks are three-dimensional (one to consider the temporal information and the other two to consider the spatial information). The spatial dimension is chosen to consider focus of attention, i.e. the size is computed so that a block covers two degrees of visual angle. The distortion measure is computed for each block by pooling the error over the channels. The actual computation of the distortion E for a given block is given by

$$E= \left (\frac {1}{N} \sum_{c=1}^N \left( \frac{1}{N_x N_y... ...x,y,y,c)\right\vert \right)^\beta \right)^{\frac{1}{\beta}},$$

where $e(x, y, t, c)$ is the masked error signal at position $(x, y)$ and time $t$ in the current block and in the channel $c$; $N_x$, $N_y$ and $N_t$ are the horizontal and vertical dimensions of the blocks; $N$ is the number of channels. $\beta$ is a constant having the value 4. The final quality measure can be expressed either using the Masked PSNR (MPSNR) as follows

$$MPSNR=10\log\frac{255^2}{E^2},$$

or can be mapped to the 5-point ITU quality scale using the equation

$$Q=\frac{5}{1+ \ \gamma E},$$

where $\gamma$ is a constant to be obtained experimentally. This can be done if the subjective quality rating $Q$ is known for a given sequence, $E$ has to be computed and then one solves for $\gamma$ from the last equation.