The Research And Implementation On Content-Based Butterfly Image Retrieval

Insects are one of the most abundant species, species-rich families on the earth. As an important suborder of Lepidoptera, butterfly is closely related with human production and life. However, at present, the identification of butterfly species is only limited to minority insect taxonomists. Except for common species, Most of butterfly species could not be identified for the common people, insect fans and agricultural technologists. Therefore, the phenomena of confusion always turned up, such as confusion of species and scientific name, confusion of beneficial and harmful insects, naming errors. This situation leads serious influences on the preservation of rare and endangered species, and the prevention of harmful species. With the development of information technology, the introduction of information technology into the butterfly species identification systems could make the task more simple, accurate and easier. Automated butterfly species identification is not only beneficial to agricultural production and the protection of species, but also of practical value to popularize the knowledge of butterfly classification.This thesis takes butterfly images as research object, collects and organizes the existing butterfly images (12 different families) in China Section. What's more, this thesis adopts content-based image retrieval technology to retrieve butterfly images. The main research contents of this thesis are as follows:(1) Based on the visual characteristics and external morphological characteristics of the butterfly images, combined with the global feature of color and shape properties of the butterfly images, the invariant moments of shape feature, the color histogram and gray histogram of the color feature have been extracted, and then these features have been applied to retrieve the butterfly images.(2) Different from the existing image feature extraction methods which start from the whole image, this thesis describes the butterfly image characteristics based on the local feature of interest points. On the basis of the local features and its relevant technology, the two local feature methods, SIFT and PCA-SIFT have been applied to retrieve butterfly image retrieval, and achieves fairly good results. (3) Mixing programming of C# and MATLAB approach has been adopted to develop the butterfly retrieval system. Firstly, color histogram, gray histogram, Hu invariant moments, SIFT and PCA-SIFT features of butterfly images are extracted. Secondly, feature matching is accomplished, during which KD-Tree and BBF technologies are applied to accelerate speed. Finally, the butterfly image retrieval system is achieved.Experimental results show the process of extract global features is simple and fast-matching, but with relatively low accuracy. Local features have higher precision rate than global features, and achieves good results.