Using Threshold Methods to Segment Highly Images beans without overlapping

Abstract

This paper presents a proposed method for segmenting
images of beans into separated objects without overlapping,
based on the area of ideal grain in (perfect) beans by
thresholding methods. The global thrshoding for the entire
image is used at first. Then local thresholding was used to
segment into sub-images, that contain overlapping between
grains. Segmentation operations for the overlapping objects into
separated objects without any overlapping depend on the
minimum value of the histogram at first in sub-image, by
repeating local thresholding until getting a value (valley between
two peaks) to split two objects. Each object is compared with the
ideal grain to put it in the result image.
The proposed method is compared with the technique
adopted in Matlab environment. By computing the average ratio
of erosion for overlapping edges objects after separating in two
methods, the proposed algorithm is 5% better than the traditional
method of 19% that means increase the accuracy by 14%,
additional clarity image (shape of grain) and the given number of
objects is (44), which is exactly too.