Superpixels using morphology for rock image segmentation

Detection and segmentation of rocks is an important first task in many applications such as geological analysis, planetary science and mining processes. Rocks are usually segmented using a variety of features such as texture, shading, shape and edges. It is easier to compute these features for rock superpixels rather than every pixel in the image. A superpixel is a group of spatially coherent pixels that form a meaningful homogeneous region, usually belonging to the same object. In this paper, we perform a comparative study of some of the current superpixel algorithms on rock images with regard to their ability to adhere to image boundaries, their speed, and their impact on rock segmentation performance. Also, we propose a new and very simple superpixel algorithm, Superpixels Using Morphology (SUM), which permutes a watershed transformation approach to efficiently generate superpixels. We show that SUM achieves a performance comparable to the recent superpixel algorithms on the rock images.