An Improved Contour Feature Extraction Method for the Image Butterfly Specimen

Abstract

Automatic recognition of butterfly species is an important and effective method to develop and utilize butterfly resources. It is beneficial not only for the timely prevention and treatment of pests in agriculture and forestry, but also for the identification of rare butterfly species in the world. At present, some specific insect species have been well differentiated in some researches. However, due to the lack of rationality of classification feature design (which fails to correlated to the biological characteristics of butterflies effectively), there are still problems of poor generalization ability, low recognition rate, and so on. Therefore, in this paper, we discussed the effects of different types of features on butterfly classification problems and improve the histograms of multiscale curvature (HoMSC) calculation method to extract the shape features of butterfly wings. To prove the effectiveness of this method, we use a weight-based k-nearest neighbor (KNN) search algorithm and 400 images of 20 butterfly species (which belong to six different families) for testing. The accuracy rate of this method reached 96%. The result suggested the improved HoMSC features can be efficient for the identification of butterfly species which are closely related.