Illumination-robust Change Detection Using Texture Based Features

We propose a change detection method which is robust against illumination change and requires little background learning as a result of using texture based features. We propose Peripheral TErnary Sign Correlation (PTESC) which is robust against illumination changes by using −1/0/1 ternary code for encoding the intensity difference between pixels in texture, and combine it with Bi-polar Radial Reach Correlation (BPRRC) which yields high detectability in a region with little texture. We show that our method detects changes with fewer false positives and false negatives under illumination changes compared with former methods.