Artificial Neural Net Assisted Edge Editing and Edge Parameter Extraction from Petrographic Images Collected With the Rotating Polarizer Stage

For any petrographic image analysis application that is aimed at the analysis of discrete objects such as mineral grains, the proper segmentation of those objects is of prime importance. However, automated edge detection algorithms are not perfect and remain problematic even in relatively clean materials such as ceramics. A practical methodology for editing edges within petrographic images is presented. The procedure uses the existing output of a standard segmentation routine as input. Because most edge detection algorithms can produce too many edges, the procedure does not find new edges; rather it examines the existing ones to identify false edges. Edges are skeletonized and converted into segments which separate two grains and join at nodal pixels. Nodal pixels connect a minimum of three separate line segments and their presence ensures that any line segment can be deleted without breaking the connectivity of any remaining boundaries. For each edge segment, an artifical neural net (ANN) evaluates differences in colour and texture parameters for the grains separated by the segment. ANN output is used to classify segments as true or false edges and can be thresholded at different levels and manually evaluated using a specialized, rapid procedure. This methodology significantly improves the speed with which edges can be edited in preparation for other studies. For each remaining individually identified edge segment, grain boundary parameters such as the length, orientation, a measure of straightness and the angle between boundary segments at nodal points can then be extracted. When combined with the size and shape parameters that can now be calculated from the correctly identified grains, this data can provide first order information about the formation or deformation of rocks.