Petrographic image classification using optimized Radial Basis Function Support Vector Machine & validation of its asymptotic behavior

Igneous rock classification using handheld specimens has been practiced regularly. The manual inspection process can be time consuming and error prone and is significantly dependent on the experience and knowledge of the observer. Hence for automated classification of Igneous rock microstructure images into two main types - Volcanic and Plutonic Igneous rocks, pattern classification approach is used. A database of locally relevant representative images belonging to both groups is considered for training as well as testing purposes. The classification into volcanic and Plutonic rocks is done on the basis of the grain size as a unique Geological textural feature. Following the identification of various fundamental statistical features such as Haralick features, and Laws Masks; a nonlinear Support Vector Machine kernel - Radial Basis Function (RBF) is used giving good classification results. It is optimized using Grid Search to offer improved results. Leave One Out Cross Validation is carried out. The Grid Search results are validated by studying asymptotic behavior of RBF SVM, plotting boundary curve, identifying underfitting and overfitting regions for the given dataset.