Classification of white rice seeds by image processing

Abstract

This research work presents the development of algorithms and programs used for classifying of white rice grains obtained from the camera by using image processing techniques. This procedure begins with improving input images, converting color images to binary images by setting the threshold value automatically by using the Otsu’s method, noise filter process image processing with structural elements that form a square, determining the grain boundary by using the Canny’s operator, determining the length of the grain by using the Euclidean’s method. Next, the length results of the grains have been obtained from the decision process based on the Rice Standards of Thailand (RST) in order to receive the type solutions of tested white rice. From the testing results for 800 images, it has been shown that in case of 500 images, the developed program provided the mean absolute error of the grain length of 0.01 mm per grain. Moreover, for testing of 300 images, the developed program provided the separation accuracy of the grains up to 99.33% in term of the RST. This can be useful in determining the price of white rice in a rice mill with farmers.