Figure 6. Combining a subject-specific and image-specific spatial bias for a better than perfect prediction.

The plots are produced as in Figure 5. The effect depends on the number of subjects that enter the bias estimation and the image category. For natural scenes, a statically significant effect (paired t-test, *** = ) can be seen when four subjects or more are used. The effect cannot be seen for urban scenes, which might be explained by the low predictive power of the subject-specific bias compared to the high predictive power of the image-specific bias on urbans.