A new local image descriptor based on local and global color features for medicinal plant images classification

Machine learning and computer vision methods are frequently applied for identifying different plant species from images in recent years. The local and global features are extracted from the image by hand-crafted algorithms such as LBP, SIFT, SURF, HOG and Hu which have demonstrated strong performance on datasets of leaves. This article propose a new method that combines local features using SIFT and bag of visual words with HOG and Hu on the two plant image datasets: Folio contains 637 images from 32 species of plants and VNPlant-255 with 255,500 samples that has a complex background. All images will be put through an image processing step, which uses image segmentation techniques. In order to evaluate the proposed approach, we conduct experiments to find a good solution. The results have shown that the SIFT + BoW - HOG - Hu techniques extracted better features than the CNN technique on a relatively small dataset and it gives 99.22% accuracy. Moreover, it also works well on medicinal plant images taken in a natural environment.