Local Binary Pattern Regrouping for Rotation Invariant Texture Classification

2. Related Work

The traditional LBP operator appeared in 1994, introduced by Ojala et al. as a local texture descriptor that works on the neighbourhood structure of pixels. It turns each pixel value of the image into a new one, which is the result of the comparison between the centre pixel and its 3*3 neighbours. This comparison is simple; each neighbour gets the value zero or one depending on its grey scale value compared to the one of the central. This is done according to the described thresholding function (S) in expression (1), where the p_c is the central pixel’s grey value and p_ni is one of the i^th neighbour’s grey value, that applies a subtraction between the two values. The eight obtained binary values are then concatenated in a clockwise order starting from the top-left neighbour to produce the new decimal value of the central pixel. When all pixels’ values of the image are transformed by the same way, the next step is to generate the histogram that describes the image and gives the possibility to analyse it:

(1)