Tomato seedling images were acquired in a two-camera image acquisition system for the seedling quality inspection. It uses two back-lit light sources, orthogonally mounted, to create seedling silhouettes for the top view and the side view image acquisitions. The images were acquired with either only back illumination or both back and side illuminations. Two adaptive thresholding techniques were selected and evaluated for segmentation accuracy. Both methods, one is based on histogram analysis and the second technique is based on the analysis of intensity and gray level gradient of local pixels, had similar performance for given lighting conditions. It is found, however, that the lighting strategy affected the image formation process. In addition, it affected the accuracy of segmentation process which separates seedling from its background.
Vladimir N. Ruzhitsky,
Peter P. Ling,
"Machine vision seedling measurement using multiple cameras", Proc. SPIE 1836, Optics in Agriculture and Forestry, (12 May 1993); doi: 10.1117/12.144026; https://doi.org/10.1117/12.144026