Fast Medicinal Leaf Retrieval Using CapsNet

Geometrical feature descriptors that describes the shape of the object are efficiently used for image feature delineation in several image recognition applications. The proposed approach implements a medicinal leaf classification system using a Bezier curve and CapsNet. A new feature vector consisting of control points (CP) of a Bezier curve and discrete fourier transform (DFT) is proposed. It also proposes a novel approach for the CP detection for feature representation. The extracted CPs are further used to find out the DFT. CPs and DFTs are further used to train the CapsNet for image classification. The CapsNet is trained to classify the input image to a particular class within a desired range of similarity. The proposed system is compared with other classification systems and the comparison reveals that the proposed system outperforms other classification systems in many parameters.