Estimation of porosity amount in building stones based on improved local binary pattern and image normalization technique

So far, various methods have been presented to detect surface defects based on image texture analysis. One of the methods that provide suitable features is the local binary pattern. According to the concept of surface defects, porosity in rock can be considered as a surface defect. In this article, a method for detecting and estimating the porosity in building stones is presented based on improved local binary pattern and image normalization technique. The presented method consists of two steps. In the training step, the one-dimensional local binary patterns descriptor is applied on the non-porous image and the base feature vector is extracted. Then the image is cropped to non-overlap windows and the feature vector is extracted separately for each window. By comparing the dissimilarity of the feature vectors with the base vector based on the logarithmic likelihood ratio, the non-porous threshold is obtained. In the detection step, the test image is windowed and windows containing porosity are identified based on the above threshold. Finally, the amount of porosity in the production defect pattern is calculated. In order to increase the detection rate, a pre-processing step is provided to normalize the images based on the single scale retinex technique. The detection rate on three types of building stones, cream travertine, orange travertine, and Tisheh'i was ۹۷.۳۳, ۹۸.۰۶, and ۹۵.۸۲, respectively. Low computational complexity and low sensitivity to noise are among other advantages of the presented method.