New Thresholding Methods for Unimodal Images of Food and Agricultural Products

Abstract

Global thresholding methods fail to segment poor contrast unimodal food and agricultural images. Many local adaptive thresholding and multi-level thresholding methods are reported in image processing journals, but there are limited studies extending them to food and agricultural images. This article presents development of Reverse Water Flow, a new local adaptive thresholding method, and Twice Otsu, a new multi-level thresholding method, to segment food and agricultural images. Reverse Water Flow method was well suited for identification of smaller objects such as 2 mm diameter holes. It reduced computational time by 61.1% compared to the previous best method. Twice Otsu method was well suited to identify larger objects. Both thresholding methods successfully segmented food and agricultural images from different imaging sources and should be extendable to other unimodal and poor contrast images. The developed methods may also facilitate further development of segmentation methods for food and agricultural applications.