自動化蝴蝶自然影像辨識系統

Abstract

本論文中，我們提出一個全新的自動化蝴蝶自然影像辨識系統，針對從真實自然環境中拍攝的全方位蝴蝶影像進行辨識。蝴蝶最主要的特徵便是其顏色；因此，在論文中，首先經由所提出的自動化區域擴散臨界值決定方法（ARGBT）以及改良的自動誤差調整K-means分群演算法（AET-K-means），自動獲得蝴蝶的主要顏色，並以此找出其所對應的多種分佈特徵。另外，於蝴蝶影像切割，我們也提出一種與使用者互動的方式，加以限制蝴蝶截取的範圍，並利用所提出的方法進行輪廓截取。而在辨識結果後，我們還提供了兩種回饋的機制，以使系統更符合使用者的需求。本系統係對26種臺灣常見的蝴蝶進行實驗，每種測試134張範例影像以及60張自然影像，而實驗結果顯示本系統能夠有效辨識蝴蝶的種類。

In this thesis, we present a novel system for recognizing butterfly from a natural image which can be taken from various shooting directions in the real scene. Color is the most important feature for butterfly recognizing. In order to extract features, we first apply the proposed methods - Automatic Region Growing Boundary Thresholding (ARGBT) and Automatic Error Threshold K-means (AET-K-means) to automatically obtain the dominant colors of butterfly, and find their corresponding distribution features. Besides, we provide an interactive method to limit the extraction area to solve the problem on object segmentation, and extract appropriate boundary by the proposed method. In addition, after the recognition process, we also support two feedback mechanisms to meet user’s expectation. Finally, experiments are conducted to demonstrate the performance of the system on the database of 26 Taiwanese common butterfly species. Each species is tested by 134 sample images and 60 natural images, and the results reveal the effectiveness for recognition.