Study on Image-based butterfly recognition using global and local features

• Main Authors: Sheng-Kai Yang, 楊盛凱
• Other Authors: Jen-Chang Liu
• Format: Others
• Language: zh-TW
• Published: 2013
• Online Access: http://ndltd.ncl.edu.tw/handle/36705906503260469814

碩士 === 國立暨南國際大學 === 資訊工程學系 === 102 === In recent years, butterfly watching tours are popularized all around Taiwan. We often see butterflies flying in the countryside, but we don’t know what kinds of butterflies they are. People usually recognize butterfly species by looking up in a butterfly handbook, but it is not convenient to carry it and hard to use for the beginners. In this thesis, content-based image retrieval is applied to recognize butterfly species by using several global and local features. The global features include HSV Color Histogram and Principal Component Analysis, and the local features include SIFT and Local Color Histogram. We conducted several experiments to find effective features for recognizing butterfly species. In this thesis, the butterfly database contains 41 common kinds of butterflies according to field researching of the Newhomeland foundation in 2011. In which three kinds of butterflies have different appearances for the males and females. As a result, the butterfly database has 44 butterfly types. In the database, each butterfly type has 10 pictures, so the butterfly database has 440 pictures. We remove the image background and normalize butterfly image before retrieving feature. The aforementioned four features were used in the experiments. The purposed Local Color Histogram has Top 1 average precision of 64.77% that is the highest rate among them.