Grey-Level Co-occurrence features for salt texture classification

Abstract

In this thesis the differences between sediment rock and salt diapir textures are discussed, and texture features are used to classify the seismic image pixels into one of two classes. Gray level co-occurrence matrices (GLCM) is a widely used statistical texture descriptor in a local window surrounding each pixel. To analyze the textures, we use the comparison of different GLCM-features to check which of them helps to separate two texture classes with the best performance. In addition, a new feature based on the calculation of the Mahalanobis distance between the average GLCMs for two classes was introduced. This feature has shown better classification results than standard GLCM features on a set of test images and can be used for salt texture segmentation in combination with other known features such as statistical, texture or features based on digital signal processing.