Automatic identification of rocks in thin sections using texture analysis

Abstract

This paper presents a test of the automatic identification of photomicrographs of rocks in thin sections using digital image processing and texture analysis. Three sets of textural measures derived respectively from the cooccurrence matrix, texture space, and texture spectrum have been used to identify six rock types (mylonite, diorite porphyry, diorite. gabbro, granite, and peridotite). Exploring only the texture characteristics of images, the average correct recognition rate reaches 89% for 58 photographs belonging to the six types. The results also show the importance of using those more discriminating texture features in the classification algorithm. For the present study, the average rate of correct classification ranges from 46% to 89% depending on the set of texture measures used. The features extracted from the texture spectrum have more discriminating performance than the conventionally used Haralick measures derived from the cooccurrence matrix.