|Kaspar, Thomas - Tom|
Submitted to: Journal of Computer Assisted Microscopy
Publication Type: Peer Reviewed Journal
Publication Acceptance Date: 9/1/1995
Publication Date: N/A
Citation: Interpretive Summary: Length and perimeter are among the most basic and useful parameters used to describe the size of objects. In many research areas of the agricultural sciences, video cameras or scanners are being used to create images of soil particles, soil structure, roots, microbes, insects, plant anatomy, animal anatomy, and worms. Often measurements of length and perimeter are needed. Currently, available image analysis programs are expensive, require lots of computer memory, and a fast microprocessor. Additionally, the algorithms for measuring length and perimeter in image analysis programs are subject to errors resulting from nonrandom object orientation within images. Because some types of objects, like plant roots, have a nonrandom orientation because of their branching pattern this results in errors in measurement. In this paper we describe an algorithm for measuring the perimeter and length of objects that does not require a fast computer with a large memory capacity and that is not affected by object orientation. We tested our new algorithm, the edge chord algorithm, against other commonly used algorithms and showed that our algorithm was more accurate and precise than the other algorithms when compared across both random and nonrandom object orientations. This algorithm will be made available to the scientific community and should allow measurement of length and perimeter of digitized objects using older, widely available microcomputers.
Technical Abstract: Length and perimeter are commonly measured parameters of digitized objects. Unfortunately, most algorithms for measuring them are not accurate for both random and nonrandom object orientations. In this paper, we present an algorithm for measuring the perimeter and length of digitized objects that draws chords along the object edge. This algorithm, ,the edge chord algorithm (ECA), does not require that the entire digitized image be held in memory and is based on rules for establishing end points for the chords. The ECA is compared with line-intercept and chain-type algorithms for perimeter and length measurements of both random and nonrandom object arrangements. For lengths of wire from 30 cm to 1500 cm cut into pieces 1 to 2 cm long and randomly arranged, the ECA algorithm underestimated length by 0.15% and was slightly less accurate, but more precise than the corner chain algorithm. For straight objects oriented parallel to each other but at different angles for each image, the intercept algorithm varied the most, whereas the most probable origin (MPO) algorithm, a chain-type algorithm, was the most precise, followed by the ECA. In the last comparison, half the objects were oriented vertically and the other half were oriented at a different angle for each image. In this comparison, the ECA was the most accurate and precise with a maximum error of 0.7% and a C.V. of 0.26%. The ECA was accurate for both random and nonrandom object arrangements and was more precise than the corner chain algorithm.