Robust plant recognition using Graph cut based flower segmentation and PHOG based feature extraction

This paper proposes an efficient computer-aided plant recognition method based on plant flower images using shape and texture features intended mainly for medical industry, botanical gardening and cosmetic industry. The target flower is segmented from the complex background using Graph cut segmentation. Shape and texture features are extracted for the segmented image. In the shape domain, a feature descriptor is developed using Pyramidal Histogram of Oriented Gradients (PHOG) that represents the image shape. It captures the distribution of intensity gradients or edge directions. Then in the texture domain, the feature descriptor is developed using Pyramidal Local Binary Pattern (PLBP). The relevant images are retrieved from the database by matching the concatenated histogram of the PHOG and PLBP feature descriptors for the given input image. Results on a database of 200 sample images belonging to different types of plants show an increased efficiency of 96%.