Clipping of variables
Based on the very relevant remarks submitted by Russel and Hoover. The two following bullets points must be investigates in order to perform safer computations.
- "compute PP T() in NS/sampling/BIL/ init .py: line 25. Negative variance is not clipped to 0."
- "After adding noise to the DCT domain, no check is done to ensure that the values are not “out of bound” in the color domain (e.g., the YCbCr domain). If the latter happens, the image would be trivially distinguishable."
Edited by Taburet Théo