Rock Particle Image Segmentation and Systems

As known, most important, and the hard part of pattern recognition for rock particles, is image segmentation. Segmentation can be divided into two steps, one is segmentation based on gray levels (called image binarization, sometimes) in which a gray level image is processed and converted into a binary image. Another is segmentation based on rock particle shapes in a binary image, in which overlapping and touching particles will be split, and over-segmented particles will be merged based on some prior knowledge such as shape and size etc. Segmentation algorithms for monochrome (gray level) images generally are based on one of two basic properties of gray-level values: similarity and discontinuity. The principal approaches in the first category are based on thresholding, region growing, and region splitting and merging. In the second category, the approach is to partition an image based on abrupt changes in gray level. The principal areas of interest within this category are detection of isolated points and detection of lines and edges in an image. The choice of segmentation of rock particle images based on similarity or discontinuity of the gray-level values depends on both developed sub-algorithms and applications. Rock particle images have their own characteristics compared to other particle images. Generally speaking, under the frontlighting illumination condition which is common case, rock particle images have the characteristics: (1) uneven background and foreground for which a simple thresholding algorithm cannot be applied to segment the images; (2) each rock particle may possess a textured surface and multiple faces, which often causes an oversegmentation problem; (3) rock particle overlapping each other, which hides parts of a particle, or causes breaks of the boundaries of rock particles; (4) touching rock particles forming a large cluster; (5) rain, snow, or much fine material making rock particle images clump together. Rock particles may be densely packed or be separated mostly on a background. The former case is more difficult to process than the latter. As well known, most systems for rock particle images were developed based on simple thresholding algorithms (some of them combined with morphological segmentation algorithm) and boundary detection algorithms. The segmentation algorithm designing is application (here, the type of rock particle images) dependent. In this chapter, the author summarize own segmentation approaches for rock particle images, they are: (1) an algorithm based on edge detection; (2) an algorithm based on region split-and-merge; (3) an adaptive thresholding algorithm; and (4) an algorithm for splitting touching particles in a binary image. O pe n A cc es s D at ab as e w w w .ite ch on lin e. co m