Processing of digital images of touching kernels by ellipse fitting

An algorithm was developed to fit ellipses to digital images of separated and touching kernels with random orientations. The algorithm was evaluated for its ability to count objects in the images and to estimate length, width, perimeter, and area of individual objects. The estimated parameters were compared with the parameters determined using conventional image processing techniques on images of physically-separated kernels. All 300 kernels were counted correctly. In some situations, it may be possible that kernels would align in such a manner that they would not be separated and counted correctly by the algorithm. The estimated size features were not significantly different from the conventionally-determined parameters at p >0.05. Keywords: machine vision, grain identification, ellipse fitting, occluding objects, separation, feature extraction.