An Effective Method for Classification of White Rice Grains Using Various Image Processing Techniques

Abstract

This chapter presents an algorithm for classifying grains of white rice by using image processing. Each image size is acquired via a digital camera. The resolution is 720 × 480 pixels. The algorithm begins with improving grain images, converting these images into binary images by using Otsu’s method, removing noise from the binary images by applying the morphological method with square structural elements, detecting each grain boundary by using the Canny operator, and determining the length of each grain by using the Euclidean method. Next, the grain length is used for classifying the rice grains according to the Rice Standards of Thailand. The testing results from processing 500 grain images; one grain per image, the algorithm provides good performance with the mean absolute error of 0.01 mm in length. For 300 grain images with some grains per image, the algorithm provides good classification with an average accuracy of 99.33 %.