植基於顯著特徵抽取技術之視訊內容分析系統

Salient Features Extraction for Video Content Analysis

吳瑞珍(Jui-Chen Wu)

指導教授 ： 陳永盛

共同指導教授 ： 謝君偉

元智大學/電機通訊學院/電機工程學系/博士(2009年)

摘要

科技的進步帶動多媒體廣泛的使用，為了有效地傳輸、儲存與檢索視訊資料，發展一套理想的物件偵測與分析顯得非常重要。為達此目的，在過去十年中，已經針對這方面提出許多相關問題。因此，若能詳盡分析平面與視訊影像中的物件，能將提供使用者極為豐富的資訊。
在本論文中，我們主要提出兩個有效顯著特徵抽取的方法，針對視訊內容中的平面中文字進行偵測與辨識，以及在交通安全系統中的車輛方向定位與資料庫中的車輛檢索。第一，我們利用“型態學基礎”的方法，此方法是利用影像中前景與背景有顯著明亮度的對比的特色，將抽取出前景的區域，對於文字偵測與辨識之中，我們將其特徵抽取之後，利用其本身的幾何特性進行文字列的偵測，接而進行影像中的車牌辨識。第二，為了將在交通安全系統中，快速偵側車輛的定位與資料庫中車輛的檢索，我們提出了一個 “特徵值顏色基礎”的方法，此方法針對某特定物體進行統計上的分析所推導得到的結果，在這個新的特徵色彩空間上，前景物像素點可以容易地與背景物的像素點作區分，即使是在一些具有光線變化的場景。抽取出候選區之後，我們將一些描述例子分別進行車子角度的偵測與車輛的檢索。實驗結果證明我們所提出的方法有不錯的效果。

關鍵字

型態學基礎 ； 文字偵測與辨識 ； 特徵值顏色基礎 ； 車子角度的偵測 ； 車輛的檢索

並列摘要

Multimedia can be widely used due to the progress of science and technology. An automatic object detection and analysis system is necessary that we can ability transmits, store and retrieve for video data. To achieve these goals, there have been many approaches proposed for detecting object for a plan or visual median. Hence, having a detailed description of video content analysis can provide rich information for user.
In this thesis, we will propose two novel salient features extraction schemes for text detection and recognition in images or video sequences, vehicle orientation analysis and vehicle retrieval from image databases. First, the morphology-based scheme can be used to find out high contrast region with their background. The method is invariant under different lighting, scaling, and viewing conditions. As a result of text often having high contrast with their background, all possible candidate regions will be extracted. Finally, the geometric properties can be used to detect text line from images. After extracting, we will recognize license plate from video sequences. Moreover, to fast and effectively analyze vehicles from image databases, we proposed the “eigen” color extraction scheme to detect possible vehicle regions from cluttered images. The model can efficiently separate foreground pixels from the cluttered images even under different lighting conditions. After extracting candidates regions, we will define some descriptors to achieve vehicle orientation analysis and vehicle retrieval system. Experimental results reveal the superior performances in text extraction, license plate recognition, vehicle orientation analysis, and vehicle retrieval.

並列關鍵字

morphology-based ； text detection and recognition ； eigen color extraction ； vehicle orientation analysis ； vehicle retrieval

參考文獻

[4]C. P. Papageorgiou, M. Oren, and T. Poggio, “A general framework for object detection,” Proceeding of 6th Computer Vision, pp. 552-562, 1998.

[5]A, Mohan, C. Papageorgiou, and T. Poggio, “Example-based object detection in images by components,” IEEE Trans. Pattern Analysis and Machine Intelligence, pp. 349-361, 2001.

[6]Z. Dekun and Y.Q. Shi, “Formatted text document data hiding robust to printing, copying and scanning,” IEEE International Symposium on Circuits and Systems, vol. 5, pp. 4971-4974, 2005.

[7]T. B. Chen, D. Dhosh, and S. Ranganath, “Video-text extraction and recognition,” IEEE Conference on TENCON, vol. 1, pp. 319-322, 2004.

[8]M. R. Lyu, J. Song, and M. Cai, “A comprehensive method for multilingual video text detection, localization, and extraction,” IEEE Transactions on Circuits and Systems for Video Technology, vol. 15, no. 2, pp. 243-255, 2005.