Algorithm for Detection of Single Isolated Human Insulin Crystals for In-Situ Microscopy

An algorithm is presented that is able to detect the regions which correspond to the single isolated human insulin crystals in a group of previously segmented foreground regions. It is based on a single nearest prototype rule, which requires the knowledge of a class prototype for each class of segmented foreground regions. Each class prototype represents a 7-dimensional mean vector of rotation, translation and scale invariant shape characteristics of several class members extracted a priori from a training set of images. An arbitrary segmented foreground region is detected as the region of an isolated human insulin crystal if the region's vector of rotation, translation and scale invariant shape characteristics is much closer to the class prototype of the regions of single isolated human insulin crystals than to the other foreground region class prototypes. The Euclidian distance is used to compute the closeness between two vectors. Experimental results with real data revealed an average processing time of 0.15 seconds/image and a detection reliability of 95 percent.