視訊物件切割技術之研究

Video Object Segmentation Using Region Matching

張鈞維(Jun-Wei Chang)

指導教授 ： 謝君偉

元智大學/工程學院/電機工程學系/碩士(2004年)

摘要

視訊影像物件切割技術在影像內涵編碼、索引或檢索方面都扮演一個非常重要的角色。因為在物件區域是由許多不同的顏色及紋理所組成，所以無法利用這些低階特徵性質直接切割出物件的區域。
本論文提出一個結合時間-空間域物件切割和互動式物件區域追蹤方式的物件切割系統，藉著定義關鍵畫面中的初始物件區域來當做參考物件區域，並且對每個畫面做時間-空間域物件切割，然後以初始物件區域的參考特徵與下一張畫面的各個切割出的區域做特徵比對，找出所有可能的區域匹配，同時以區域相鄰關係圖來表示區域之間的相對位置與特徵結構，做為後面畫面的特徵比對形式，接下來利用區域合併和動態輪廓的方式將剩下的符合條件的區域標記成物件區域，使物件從影片中切割出來。本論文所提供的實驗結果將證明本方法有較好的表現。

關鍵字

時間-空間域物件切割 ； 特徵擷取 ； 區域追蹤 ； 特徵比對 ； 區域相鄰關係圖 ； 動態輪廓

並列摘要

Video segmentation and object extraction play a very important role in supporting content-based image coding, indexing, and retrieval. Since a real object have very different visual features (like colors, textures, and shapes) when it appears in a video, the work to effectively semantic objects from videos has become very challenging.
This thesis proposes a semantic video object segmentation system which combines spatial-temporal video segmentation and interactive region tracking together to extract important semantic objects from videos. At beginning, the paper uses a semi-automatic technique to extract and define a reference object from an initial video frame. Then, a technique of spatial segmentation is used to segment all next frames to several non-homogenous regions. Further, according to temporal information of each segmented region, we can find all possible region matching pairs by matching the reference object to all segmented regions. These matching pairs can provide important information to build a region adjacency graph (RAG) which can well record the relative relations between the reference object and each segmented regions. Through region grouping and relation checking from this RAG, different regions will be merged and associated together to form a meaningful object. Since this object’s boundaries may be occluded by other regions, a technique of active contour tracking is further used to refine this desired object more accurately. Experimental results have proved the superiority of the proposed method in object segmentation.

並列關鍵字

spatio-temporal segmentation ； feature extraction ； region tracking ； feature matching ； region adjacency graph ； active contour

參考文獻

[1] H. J. Zhang, C. Y. Low, Y. Gong, and S. W. Smoliar, “Video Parsing Using Compressed Data,” Proc. IS&T/SPIE Image and Video Processing II, pp.142-149, February 1994.

[2] B. C. Song and J. B. Ra, “Automatic Shot Change Detection Algorithm Using Multi-stage Clustering for MPEG Compressed Videos,” Journal of Visual Communication and Image Representation, vol. 12, no. 3, pp.364-385, Sep. 2001.

[4] H. J. Zhang, A. Kankanhalli, and S. W. Smoliar, “Automatic Partitioning of Full-Motion Video,” ACM Multimedia System, vol. 1, no.1, pp. 10-28, 1993.

[5] A. Nagasaka and Y. Tanaka, “Automatic Video Indexing and Full-Video Search for Object Appearances,” Visual Database Systems II, pp. 113-127, 1991.

[6] M. Kim, J. G. Choi, D. Kim, H. Lee, M. H. Lee, C. Ahn, and Y. S. Ho, “A VOP Generation Tool: Automatic Segmentation of Moving Objects in Image Sequences Based on Spatio-Temporal Information,” IEEE Trans. Circuits and Syst. Video Technol., vol. 9, no. 8, Dec. 1999.