- 學位論文
以顏色與灰階梯度特徵為基礎的昆蟲影像辨識系統
Insect Image Recognition System Based on Color and Intensity Gradient
摘要
昆蟲是世界上最多的生物之一，分佈廣泛且種類眾多，因此有著許多不同的樣貌，日常中處處可以看見它們的蹤跡，非常適合做為教學用的素材。因此一些學術單位或是學校常利用昆蟲數位影像建置昆蟲資訊網站，提供大眾或是學生一個學習的管道，另外提供了關鍵字查詢和外型查詢等檢索功能。但是昆蟲種類眾多，外型特徵複雜，容易造成使用者的混淆，因此學生不容易能檢索到想要的昆蟲資訊。 有學者就利用影像辨識技術建置昆蟲的辨識系統，讓使用者可以較簡單查詢到昆蟲資訊，但是目前的昆蟲影像辨識系統，對於昆蟲的紋理特徵描述仍不足夠，以至於容易產生錯誤判斷，因此本研究以梯度特徵開發出一套可以細膩比對物件紋理差異的方法，另外將區域分佈的概念導入顏色特徵的描述方式，使顏色特徵的描述除了顏色差異外，還包含顏色分佈，顏色差異的計算上，採用了CIE1976色差公式進行比較。 本研究藉著改進影像辨識對於物件特徵的描述，達到提升系統對昆蟲的描述力與判斷力。
並列摘要
Insects are the most popular creatures in the world.Due to their various appearances, they are very suitable for using as teaching materials.Many schools and academic units had set up web sites to provide insect images and information.Some websites possessed the indexing functions using keywords or shapes.However, since the insects were composed of many species and with complex shapes, the users were easily confused and can not retrieve the desired insect information. Some researchers tried to incorporate the image recognition mechanism,thus, the users only needed to provide the image of the interested insect.The system would recognize the insect and retrieve the relevant information.One important factor affecting the recognition rate was the texture of the insect.In this thesis,we would develop a gradient-based feature descriptor which can differentiate the fine details of the textures of the insects.CIE1976 color difference formula was applied to compare the color difference.In addition, we compute and compare the color distributions region by region to preserve the spatial information.Testing with 455 folky Coleoptera insects, our system can achieved 94.28% recognition rate.