Fragmented plant leaf recognition: Bag-of-features, fuzzy-color and edge-texture histogram descriptors with multi-layer perceptron

Plants species recognition is one of the most important research topics in the biological sciences. Although leaves are convenient markers for identification, a major drawback is that they are prone to be damaged easily by various environmental and biological factors. The proposed research work aimed to tackle this situation by proposing a leaf recognition system that can specifically handle fragmented leaf images. As leaf images are fragmented they can not be recognize based on shape features. Here a novel approach is proposed by using the combination of fuzzy color and edge-texture histogram in order to recognize fragmented leaf images. First, the dataset leaf images that are similar to the query fragmented leaf image is identified by using bag-of feature. Then, the combined feature is used to generate the feature vector. Since fragmented leaves provide less information, this work also attempted to derive a fragment size threshold beyond which results become unpredictable, and whether such thresholds are universal or vary depending on other factors. The efficacy of the proposed method was studied using a multi-layer-perceptron classifier. As there is no public database of the fragmented image, a method was designed to create the reproducible each fragmented leaf image from the whole corresponding one.