Flow-based Segmentation of Seismic Data
Master Degree Thesis
by Kari Ringdal
supervised by Daniel Patel
Abstract
This thesis presents an image processing method for identifying separation layers in seismic 3D reflection volumes. This is done by applying techniques from flow visualization and using GPU acceleration. In geology sound waves are used for exploring the earth beneath the surface. The resulting seismic reflection data gives us a representation of sedimentary rocks. Analyzing sedimentary rocks and their layering can reveal major historical events, such as earth crust movement, climate change or evolutionary change. Sedimentary rocks are also important as a source of natural resources like coal, fossil fuels, drinking water and ores. The first step in analyzing seismic reflection data is to find the borders between sedimentary units that originated at diffrent times. This thesis presents a technique for detecting separating borders in 3D seismic data. Layers belonging to different units can not always be detected by looking in a local neighborhood. Our presented technique avoids the shortcoming of existing methods that work on a local scale by addressing the data globally. We utilize methods from the fields of flow visualization and image processing. We describe a border detection algorithm, as well as a general programming pipeline for preprocessing the data on the graphics card. Our GPU processing supports fast filtering of the data and a real-time update of the viewed volume slice when parameters are adjusted.