Semi-automated characterization of thin-section petrographic images

Abstract

This thesis represents the continuation of work on PetrograFX, an automated image analysis toolkit for petrographic image analysis. These types of images are used in the petroleum industry to provide valuable information, however, the retrieval of data from these images is time consuming and prone to operator bias. An integrated solution that combines a number of basic image processing concepts, each tailored towards segmenting a particular type of grain, is developed to automate this process. Specifically, an attempt is made to replicate the methodology and analysis carried out by core laboratories, which typically place more emphasis on overall 'interpretation of the image rather than just the measurement of the porosity and quartz grain distribution. This requires a solid treatment of the geological background to ensure the data being collected will be useful. Due to their complex nature there will be regions within these images that are unidentifiable. This approach necessitates a classification routine to eliminate objects once they have been segmented to ensure that they are unaffected by subsequent routines. To provide a quick and objective assessment segmentation performance an automated accuracy routine is presented.