Color CENTRIST：用於場景分類的顏色描述子

Color CENTRIST: A Color Descriptor For Scene Categorization

陳志豪(Chih-Hao Chen)

指導教授 ： 朱威達

國立中正大學/工學院/資訊工程學系/碩士(2012年)

摘要

對於媒體分析來說，圖片的場景訊息可以帶來豐富的資訊，因此場景分群被應用在許多地方並且擔任重要的角色。目前與場景分群相關的研究多專注於使用灰階圖，使用藉由像素的灰階值計算出的方向傾斜度(oriented gradient)做為區域描述子(local descriptor)，並將這些區域描述子組成袋字模型(bag of word model)來描述一張場景圖片。然而在這種情況下，視覺辭彙(visual word)可能由不同場景中的不同物件產生，導致這些辭彙在辨別場景上的能力不佳。除此之外，袋字模型在其運算上所需花費的時間相當高。在本篇論文中，我們提出了一套快速的場景分群系統來解決上述的問題。
我們希望運用彩色圖豐富的色彩資訊進行場景的分群。我們設計了一個方法，將彩色資訊融合到Census 轉換統計直方圖(CENsus Transform hISTogram, CENTRIST)的架構中，CENTRIST是目前用於場景分析，最新的descriptor；這種全域描述子(global descriptor)與區域描述子不同的是能夠完整紀錄圖片區塊的整體結構，而且不會被太過於細節的紋理資訊所干擾。此種特性非常適合用在場景分群。基於CENTRIST的架構下，我們提出Color CENTRIST，不僅運用CENTRIST原本既有的優點，並且充分的利用豐富的顏色資訊。Color CENTRIST除了藉著由顏色索引值(color index)算出來的傾斜度(gradient)描述整體的形狀，而且還能夠描述圖片每個區域在像素之間顏色的變化。藉由顏色的資訊，更能夠精確地將場景分群。
透過在各個不同的資料庫進行實驗，證明Color CENTRIST不僅容易實作，而且確實達到比CENTRIST更好的效能。使用顏色資訊確實有益於場景分群上。

關鍵字

場景分群 ； Census 轉換統計直方圖 ； 彩色描述子

並列摘要

Scene categorization acts as an essential part in many applications since scene type of an image provides abundant information for media analysis. Most works about scene categorization target on gray images, and rely on oriented gradient calculated based on intensity values as local descriptors. With these descriptors, the bag of word model is used in describing scene images. However, a visual word may be generated from different objects in various categories, and discriminative capability of visual words may hence decrease. On the other hand, exhaustive computation makes processes inefficient. In this thesis, we propose a fast scene categorization system to solve the problems mentioned above.
We would like to study scene categorization for color images. We devise a new visual descriptor that incorporates color information into the framework of CENsus TRansform hISTogram (CENTRIST), a state-of-the-art visual descriptor for scene categorization. CENTRIST mainly encodes the structural properties within an image and suppresses detailed textural information. It is suitable to place and scene recognition task. Based on CENTRIST, we devise a new visual descriptor, i.e., color CENTRIST, that incorporates the advantage of CENTRIST and color information. The newly proposed color CENTRIST descriptor describes global shape information by not only gradient derived from intensity values but also color variations between pixels in local image patches. With color information, scenes of images can be effectively categorized.
Through extensive evaluations on various datasets, we demonstrate that the color CENTRIST descriptor is not only easily to be implemented, but also reliably achieves performance over that of CENTRIST. Considering color information indeed benefits scene categorization.

並列關鍵字

Census transform histogram ； Scene categorization ； Color descriptor

參考文獻

[1] J. Wu and J.M. Rehg, 2011. CENTRIST: a visual descriptor for scene categorization. IEEE Transactions on Pattern Analysis and Machine Intelligence, vol. 33, no. 8, pp. 1489-1501.

[2] J. Sivic and A. Zisserman, 2003. Video Google: a text retrieval approach to object matching in videos. Proceedings of IEEE International Conference on Computer Vision, vol. 2, pp. 1470-1477.

[3] J. B. MacQueen, 1967. Some Methods for Classification and Analysis of Multivariate Observations. Berkeley Symposium on Mathematical Statistics and Probability, pp. 281-297.

[4] C. C. Chang and C. J. Lin. 2011. LIBSVM: a library for support vector machines. ACM Transactions on Intelligent Systems and Technology, vol. 2, no. 3. pp. 27. Software available at http://www.csie.ntu.edu.tw/~cjlin/libsvm.

[5] A. Oliva and A. Torralba. 2001. Modeling the shape of the scene: a holistic representation of the spatial envelope. International Journal of Computer Vision, vol. 42, no. 3, pp. 145-175.