Once the images have been acquired and scaled, the next step is for Split-Desktop to delineate the individual rock fragments in each of the images. A number of new features have been added that increase both the accuracy and the user-friendliness of the fragment delineation step.

First of all, some new preprocessing algorithms have been added to better take into account unwanted shadows in the images. Secondly, an automatic thresholding algorithm has been added. Previously the optimum threshold for each image was determined by the user, and required a certain amount of expertise by the user. The auto thresholding routine significantly reduces the training and production time to use the Split-Desktop program.

After preprocessing and auto thresholding, the Split-Desktop program automatically delineates the fragments using a set of algorithms based on the following 4 steps: gradient filter, shadow convexity analysis, Split algorithm, and Watershed algorithm. Details of these steps are described in Kemeny (1994) and Girdner et al. (1996).

The result of the automatic delineation is a binary image (2 graylevels, black and white) that contains white particles and a black background. Figure 6 is the binary image that results from the delineation of the muck pile image shown in Figure 3 (some editing has also been performed as described in the next section). The black areas in these images contain fine material too small to delineate in addition to the unfilled air space between particles. This black area is very important in estimating the amount of fines, as described below.

In most muck pile images and in many images from other sources such as haul trucks or leach piles, there are instances when the automatic delineation algorithms in Split-Desktop will not delineate the fragments properly. This may be due to situations where the lighting is poor, there is an abundance of fines in the image, and the image quality is low or other reasons. In these cases, the binary file containing the delineated fragments needs to be edited using hand editing tools in the program.

There are three common cases where minor editing is needed. First of all, if there are large patches of fines in the image, Split-Desktop sometimes mistakes these patches as a single large fragment. Secondly, if there is excessive "noise" on a fragment (due to bedding, rock texture, etc.), the Split-Desktop program may split this fragment into a number of smaller fragments. Thirdly, some of the delineated particles are neither rock fragments nor fines, such as the balls in Figure 3.

Split-Desktop has built in editing capabilities to handle the situations described above. The program first makes a stack of images, where one file in the stack is the delineated image and the other file in the stack is the original grayscale image. The user can quickly toggle between the original and delineated images to determine which parts of the image need editing. Three kinds of editing are most common: paint bucket filling of fines, erasing unwanted delineations, and identifying non-rock features. In most cases the images can be edited by a skilled user in less than 5 minutes.