Improving pore type identification from thin section images using an integrated fuzzy fusion of multiple classifiers

A semi-automatic algorithm for identification of five pore types in carbonate rocks of a gas field, including inter-particle, intra-particle, oomoldic, biomoldic, and vuggy, from thin section images is presented. The proposed algorithm involves four main steps. Firstly, a color-based image segmentation procedure is carried out using K-means clustering algorithm to separate regions corresponding to the considered pores. Secondly, six geometrical shape parameters of 384 pores are extracted from each segmented region to obtain the distinctive features for each pore type. Three classifiers (i.e. kNN, RBF, and SVM) are considered for classification to identify the type and percentage of interested porosities. Experimental results show that SVM with polynomial kernel function yields the highest accuracy and can effectively identify the pores with average accuracy of 94.4% whereas the kNN gives the lowest accuracy. The final step include combination of outputs of three employed classifiers that are trained separately on same dataset to recognize each pore space using Fuzzy Sugeno Integral (FSI) method. As expected, the fuzzy fusion of single classifiers improves the results of classification up to 9.4%, which is a considerable improvement in the classification from the stand point of petroleum geology.