Automatic classification of non-touching cereal grains in digital images using limited morphological and color features

Classification of cereal grains, namely; barley, oat, rye and wheat (Canada Western Amber Durum (CWAD) and Canada Western Red Spring (CWRS)) was performed using morphological and color features. Grain image boundary contours were extracted from the digital images of kernels, expressed as chain-coded points and then approximated by 13 elliptic Fourier coefficients. After normalization of the rotation and starting point of the contours, symmetrical standard coefficients were determined. The symmetrical Fourier index (S-FX) of individual kernels was calculated from the product of the sum of absolute symmetrical coefficients and the circularity (roundness) index. Three geometric features, namely; aspect ratio (AR), major diameter (M-D) and roundness (C-eq) were determined using ellipse fitting and Green's transformation of curve integrals, respectively. The morphological classification model was defined using S-FX, AR, M-D, and C-eq. The color classification model was defined using color indices of individual kernels, which were calculated from the RGB color values of their images. The classification accuracies of different models were evaluated and compared. The combined model defined by morphological and color features achieved a classification accuracy of 98.5% for barley, 99.97% for CWRS, 99.93% for oat, and 100% for rye and CWAD. (C) 2012 Elsevier B.V. All rights reserved.