AbstractThirty years of microtextural study of quartz grain form has revealed many different textural types and shapes. In conjunction with other sedimentological methods, surface textures and shapes have been assigned to quartz grains from a variety of depositional settings. Many different methods have been devised to semi-quantify these variables and identify sedimentologically meaningful surface features. However, ambiguity and controversy about feature recognition has questioned the method's credibility. Recent attention has turned to new methods of texture and shape analysis using computer image processing techniques. This approach was suggested as a framework for eliminating human subjectivity, to speed up processing time and provide quantitative methods of analysis which were reproducable and available to a wide variety of researchers.
Quartz grain images from a Scanning Electron Microscope were 'frame grabbed' and converted into digital images using a P.C. based image analysis system. The use of digital imagery allowed a number of pre-existing textural and shape algorithms to be applied for the first time. Three differing quartz grain textural types were used (Beach, Desert and Crushed Quartz), employing three hundred grains per sample. A new quantitative method for describing quartz grain shape using Fourier Descriptors removed problems inherent in earlier Fourier implementations, particularly the difficulty in dealing with re-entrant values generated by irregular shaped grains. In addition, a variety of textural algorithms (Autocorrelation, 2-D Fourier Transforms, Textural Units, Edge textures and Single Gray level Dependency Matrices) were applied to quartz grain surfaces. While all algorithms were successful in separating different textural types, edge textures proved to be the most discriminating and potentially the best individual method of analysis.
This departure from established methods has provided exciting new methodologies for quantitative analysis of quartz grain form. Statistical analysis confirmed that textural quantification was possible and when several textural methods were used in conjunction with shape analysis, provided a powerful environment for the study of quartz grain morphology.
Discriminant analysis was able to separate grain types, with an average 80% success rate. The use of a neural network, employing a back propagation paradigm, proved successful as a method for automatic quartz grain texture recognition.
|Date of Award||1992|