Understanding of the view (scene) and 3D objectrecognition is one of the magnificent challenges in computervision. A wide variety of techniques and goals, such as structurefrom motion, optical flow, stereo, edge detection, andsegmentation, can be viewed as subtasks in scene understandingand object recognition. Many methods can be applied by previousinvestigators. On the contrary, this research is focused on highlevelrepresentation for scenery and objects, especially physicalrepresentation recognize 3D view of the underlying image. Thisstudy aims to answer the following questions:• How to relate 2D image with a 3D scene, and how we can takeadvantage of the relationship perspective?• How does the physical scene space can be modeled, and how toestimate the space scene of an image?• How to represent and recognize objects in a way that is robust tochanges in viewpoint?• How can use the knowledge and perspective of the scene toimprove the recognition space, or vice versa?In this study, carried out the stages of development of the plant recognition system based on 3D stereo images leaves, namely: image enhancement and segmentation, stereo correspondence, disparity map calculation and depth maps, feature extraction using Gray Level Coocurence Matrix, and classification using Euclidian distance. The results obtained in this study indicate that the recognition accuracy of the plant with the highest 3D image of the leaf is 83.3% to recognize 3 varieties of plants. While to recognize 9 varieties of plants obtained low accuracy. The low accuracy is due to the quality of the disparity and depth maps are possible for further research.