應用在視訊檢索之智慧型人臉偵測系統

An Intelligent Face Detection System for Video Retrieval

羅啟宏(Chi-Hung Lo)

指導教授 ： 陳淑媛

共同指導教授 ： 王照明(Dr. Chao-Ming Wang)

元智大學/資訊學院/資訊工程學系/碩士(2006年)

https://doi.org/10.6838/YZU.2006.00010

摘要

有關於檢索的應用，最早之前，是以文字與圖形來作為檢索的應用。但是隨著現代科技的進步，網路的頻寬也日漸增加，再加上使用者的需求更多元化，許多人紛紛發表出視訊來作為檢索的技術，包括高速網路影像查詢、虛擬博物館、專利圖查詢等。本論文的目的，即提出一個利用人臉偵測的技術來記錄畫面中的人物出現片段的系統，以方便使用者迅速擷取找尋人物的片段，以節省寶貴的時間。
本論文所提方法分成三大部份，影像前處理、人臉偵測及視訊檢索。在影像前處理階段，首先利用膚色偵測，將可能的候選區塊擷取出來，接著進行消除雜訊等前處理動作。在人臉部偵測階段，則先判斷每個區塊是否有眼睛與嘴巴的特徵。最後，在視訊檢索階段，則定義一個資料結構，其中包含了編號、啟始訊框、結束訊框、訊框長度以及正面訊框。每當畫面中判斷有人物出現時，則開始記錄該人物進入畫面的啟始時間，並且運用人物追踨的方法來判斷是否該移動者是屬於同一人，當偵測到該人物離開畫面時，則記錄結束時間。同時，系統會自動顯示出在此期間中出現最正面臉孔的訊框供使用者觀看。如此，可獲得一檢索檔，其內容即為該人物出現此畫面的片段。因此，一個完整的視訊檔案，透過本論文所提方法，可以拆解成許多人物出現的片段，以供使用者方便檢索。實驗結果顯示所提方法確實有效可行。

關鍵字

人臉偵測系統、智慧型搜尋、圖形辨識、顏色資訊、視訊檢索

並列摘要

For a long time, words and patterns were applied for retrieval. Since technology is rapidly changing the capabilities of computer science and the network bandwidth is increasing more and more, it makes users want to retrieve images using other methods. For example, people may try to use video to retrieve image on high speed network, virtual content of museum, and patent of pattern, and so on. In this study, face detection approach is proposed for video retrieval, so as to provide convenient environment for users to rapidly find someone in a video stream.
The proposed method is based on color information by converting RGB into HSI color space to extract skin region with specific colors. All the skin regions are then pre-processed to eliminate noise, fill up hole, and so on. Finally, in the stage of video retrieval, retrieval structure is proposed in this study. There are some fields in this structure: “number of records”, “starting frame”, “ending frame”, “length of frames”, and “frontal face frames”. When someone are detected by our system in certain frames, current number of frame is recorded as “starting frame” and the proposed tracing approach is performed to judge whether the moving object is belonging to the same one. If the tracked people disappear in certain frame, the current number of frame is recorded as “ending frame”. Finally, the length of frames is automatically computed and the most proper frontal face is displayed on the user interface.
As a summary, retrieval records related to someone appearing in a video stream are maintained, so as to provide convenient information for user to rapidly find someone in the video stream. The feasibility and practicability of the proposed method are demonstrated by various experiments.

並列關鍵字

Face detection system, Intelligent searching agent, Pattern recognition, Color information, Video retrieval

參考文獻

[12] S.K. Singh, D.S. Chauhan, M. Vatsa, and R. Singh, “A robust skin color based face detection algorithm,” Tamkang Journal of Science and Engineering, vol. 6, no. 4, pp. 227-234, 2004.

[1]T.K. Leung, M.C. Burl, and P. Perona, “Finding faces in cluttered scenes using random labeled graph matching,” Proc. Fifth IEEE Int’l Conf. Computer Vision, pp. 637-644, 1995.

[2] K.C. Yow and R. Cipolla, “Feature-based human face detection,” Image and Vision Computing, vol. 15, no. 9, pp. 713-735, 1997.

[3] C.H. Lin and K.C. Fan, “Triangle-based approach to the detection of human face,” Pattern Recognition, vol. 34, pp. 1271-1284 , 2001.

[4] T.S. Jebara, K. Russell, and A. Pentland, “Mixtures of eigenfeatures for real-time structure from texture,” Proc. Sixth IEEE Int’l Conf. Computer Vision, pp. 128-135, 1998.