利用多實例學習法為基礎之鳥類影像檢索系統探討

Bird Image Retrieval System Based on the Multiple Instance Learning

莊惠傑(Huei-Chien Chuang)

指導教授 ： 蔡偉和

國立臺北科技大學/電資學院/電腦與通訊研究所/碩士(2013年)

摘要

近年來，由於智慧型手持裝置技術的演進，取得影像變得相對容易，使得影像識別蓬勃發展，且被廣泛運用於日常生活中，如：鳥類影像辨識、指紋辨識、車牌辨識系統等。
在鳥類影像檢索上，普羅大眾多半利用圖鑑等科普書籍，藉比對書籍上之圖片以及觀察到之鳥類進行比對及辨別，但此辨識動作需要一定的經驗與知識基礎，難免讓初學者失去興趣，造成利用鳥類作為科普材料的障礙。為克服此障礙，本論文對於台灣常見鳥類影像，提出一個自動辨別鳥類影像的系統。其中的鳥類影像可能在許多包含不同背景所拍下，因此，系統必須識別哪些區塊內含有鳥類影像，在鳥類影像擁有物件類別多樣性、影像組成複雜度高、資料變異度高、物件出現頻率不均的因素下，使得系統在鳥類影像辨識的技術開發上，非常具有挑戰性。
本論文從網路收集了大台北地區常見的20種鳥類共440個影像樣本進行實驗，並對此樣本集使用了賈柏濾波器作為鳥類影像特徵擷取，而後利用這些特徵值擷取資料庫，透過Multiple-Instance Decision Based Neural Networks(MI-DBNN)演算法，進而識別出鳥種。

關鍵字

影像檢索 ； 賈柏濾波器 ； 多實例學習

並列摘要

Nowadays, it is getting easier to obtain images with smart-carrying device, which makes image identifying become popular and being applied in common use as fingerprint, car license and bird-shaped recognition.
Mostly, people use illustrates science popular books to identify the birds they see in lives and pictures. However, to identify birds with books need experience and research which would let beginners struggle with lack of knowledge. To deal with the problems, a device system was designed according to the theory discussed in this article. In this system, it will automatically identify the appearance of birds then provide information. The images of birds were collected with different surrounding, then the system need to define the area with exteriors of birds. There are challenges of the technology of exploiting the system - the diversity of the species of birds; the complexity of the construction of the images; also, the unstable possibility of collecting specimen, which are needed to conquer in this article.
This article is based on the experiment with 440 images of birds with 20 specious in Taipei as sample group. The sample group was dealt with Gabor filter to derive the contour values of birds and use the values as data base then calculate with MI-DBNN method to define the specious of birds.

並列關鍵字

Image Retrieval ； Gabor Filter ； Multiple Instance Learning

參考文獻

6. B. G. Buchanan and T. M. Mitchell, Model-directed leaning of production rules, Academic Press, 1978.

8. H. Xu and H. Yaguchi, “Visual evaluation at scale of threshold to suprathreshold color different,” Color Research and Application, vol. 30, no. 3, pp. 198-208, 2005.

9. J. B. MacQueen, “Some Methods for classification and Analysis of Multivariate Observations,” Proceedings of 5-th Berkeley Symposium on Mathematical Statics and Probability, Berkeley, University of California Press, 1:281-297, 1967.

10. T. G. Dietterich, R. H. Lathrop, and T. Lozano-P’erez, “Solving the multiple-instance problem with axis-parallel rectangles,” Artificial Intelligence, vol.89, no.1–2, 1997, pp.31–71.

16. Thomas Gartner, Peter A. Flach, Adam Kowalczyk and Alex J. Smola, “Multi-instance kernels,” in Proceedings of 19th International Conference on Machine Learning, 2002, pp. 179–186.