Research on Flower Image Classification Using Sum-Product Networks

Abstract

Traditional stretch-based preprocessing method in image classification is unable to meet practical needs due to image distortion and information destruction of original structures. This paper presents a novel approach based on Sum-Product Networks (SPNs) for flower image classification. With SPNs, an image preprocessing method maintaining aspect ratio is used and an efficient feature extraction method is then applied. We performed experiments on 17 classes of flower images to examine the efficiency of the method we proposed.