Mobile image analysis for seed phenotyping
Metadata[+] Show full item record
[ACCESS RESTRICTED TO THE UNIVERSITY OF MISSOURI AT AUTHOR'S REQUEST.] Shape and size are among the most important seed characteristics measurements. This work presents an Android-based smartphone and tablet application named MUSeed that utilizes image processing techniques for seed morphometry. Unlike most of the existing tools, MUSeed does not impose restrictions on arrangement of seeds in an image since it is capable of handling touching seed instances. Firstly, RGB color space is converted to RG-chromaticity space to reduce the influence of shadows and illumination variations. Then K-means is used to segment the seeds from the background. To split touching seeds in the image, a Modified Marker-Controlled Watershed (MMCW) algorithm and Concave Point Analysis (CPA) are developed. Finally, the most appropriate result from two methods is selected by a fitness function and each seed is then labelled and measured. The proposed app used Android Studio as the Android application IDE. The program was implemented in Java with the help of OpenCV computer vision library. MUSeed was benchmarked against several similar tools on 7 cultivars of American elderberry (Sambucus nigra subsp. canadensis.) seeds: Bob Gordon, Marge, Ocoee, Ozark, Ozone, Wyldewood, and York.