Figure 5.

Error (NRMSE) for one example subject using Ridge Regression with 5-fold cross-validation, training with on/off values and testing with all values, for (A) increasing blur, where the number on the x-axis corresponds to the kernel size of the Gaussian blurring; (B) decreasing contrast (α), where 1 is full contrast and 0.2 corresponds to 20% of the original contrast value; and (C) changing brightness (β), where 0 corresponds to the original brightness value, negative values correspond to underexposure of the image and positive values to overexposure.