where is the average of pixel luminance of region in one
view. The value of can be obtained by minimizing the mean
squared error between all subjective values in the luminance
masking experiment (as shown in Fig. 3(a)) and the predicted
values according to the BJND model indicated by (6). The root
mean squared error (RMSE) scores under different values are
listed in Table I, which suggests the value for binocular noise
combination is within the range of [1.0, 1.5]. In this letter, we
use which tends to yield the best fitting performance.
The initial luminance-masking-based BJND profile is shown as
the surface in Fig. 3(c).