Developing machine vision algorithms that can correctly segment plant stems can significantly impact plant agriculture in greenhouses. If the stem can be accurately segmented, then other operations like growth monitoring, disease detection, and robotic de-leafing can be completed autonomously with minimal human intervention. However, segmenting plant stems in visual images of plants in greenhouses is a challenging task. Plants are highly occluded, overlapping, and grow in different directions. This thesis investigates this problem using three approaches. The first is based on modeling the plant as a non-rigid object: Active Shape Model (ASM). The second is based on machine learning: the Hough-forest technique. The last is based on using heuristic image processing algorithm: Vine search. The three approaches were tested on a benchmark dataset consisting of 100 images collected from commercial greenhouse operations in southern Ontario, Canada. Results show that while ASM handles deformable objects the best, it required a lot of processing time to achieve good results. Hough-forest outperformed other algorithms. It had the highest segmentation accuracy (99.6%), and it required the shortest processing time.