Experimentally Derived Weighting Function in Spatial Frequency Domain for Evaluating Still Pictures

Abstract

A “weighting function” is defined for evaluating monochrome still-picture quality in the spatial frequency domain. The weighting values of the function can be determined from the relationship between the MOS (Mean Opinion Score) and the spatial-frequency-selective noise level. A hundred and six weighting values are calculated in the spatial frequency dommain of 256 (TV lines) x 256 (TV lines). When the ratio of viewing distance to picture height is 7, the weighting value at 3 (cpd, cycle per degree) shows maximum value. The function is isotropic in the region from 3 (cpd) to 8 (cpd) and anisotropic at over 10 (cpd).
Using the weighting function, a new objective measure named WSNR (Weighted Signal-to-Noise Ratio) is also defined and applied to pictures degraded artificially. The picture quality is degraded by adding random noise, by passing through a two-dimensional low-pass filter, or by distorting blocks. Experiments using different portrait pictures show good correlation coefficients between the MOS and the WSNR, 0.93-0.97.