Authentication of rice by three-sided image analysis of kernels using two mirrors

Abstract

The authentication of rice by imaging three sides (top, side, and front) of a rice kernel using two mirrors was attempted. Nine image characteristics (area, aspect ratio, minimum diameter, maximum diameter, perimeter, and red, green, and blue values) for each side of a rice kernel were measured. Japonica type rices were used for the study; a total of 89 rice samples were used: 40 samples harvested in Korea, 33 in America, 5 in Australia, and 11 in China. Image data collected from 105 kernels per rice sample were used a calibration data set to develop a model for quadratic discriminant analysis. In overall pairwise trials, on average, 85.8-94.5% correct identification of the cultivated country was obtained, depending on the number of image characteristics used for the discriminant model. For the effective discrimination of the cultivated country, 13 image characteristics were selected. As a test sample set, 20 kernels per sample were used. A model for discriminant analysis with 13 image characteristics yielded an average of 95.0 and 97.5% accuracy for calibration and test sample set, respectively.