Image based plant phenotyping plays a vital role in productive and sustainable agriculture. It is used to record plant growth, chlorophyll fluorescence, yield, width and tallness of plants and leaf area, frequently and accurately. Among these characteristics, plant growth is an important characteristic to be analyzed, which directly depends on leaf count. Taking benign conditions of quick advancement in computer vision and image processing algorithms, a new method is proposed to extract the leaves from complex background and to count the number of leaves. The proposed method has two stages. In the first stage, leaf region is separated from the background, using graph based method. In the second stage, leaves are counted by using Circular Hough Transform (CHT). The proposed method is experimented with Leaf Segmentation Challenge (LSC) benchmark datasets. The proposed method achieves the Dice score of 93.2% and FBD of 94.3%, which are higher when compared with the existing recent relevant methods.