Automated image analysis for petrographic image assessments

Abstract

In this thesis, the algorithms developed for an automated image analysis toolkit called PetrograFX for petrographic image assessments, particularly thin section images, are presented. These algorithms perform two main functions, porosity determination and quartz grain measurements. For porosity determination, the pore space is segmented using a seeded region growing scheme in color space where the seeds are generated automatically based on the absolute R - B differential image. The porosity is then derived by pixel-counting to identify the pore space regions. For quartz grain measurements, adaptive thresholding is applied to make the system robust to the color variations in the entire image for the segmentation of the quartz grains. Median filtering and blob analysis are used to remove lines of fluid inclusions, which appear as black speckles and spots, on the quartz grains before the subsequent measurement operations are performed. The distance transformation and watershed transformation are then performed to separate connected objects. A modified watershed transformation is developed to eliminate false watersheds based on the physical nature of quartz grains. Finally, the grain are characterized in terms of NSD, which is the nominal sectional diameter, NSD distribution and sorting.