利用3D KINECT影像做視訊監控應用上之隱私保護與秘密隱藏

Abstract

隨著立體電腦視覺技術的快速發展，3D影像設備越來越普及，可在各種條件下讀取3D多媒體資訊。微軟的KINECT設備即是一著名的例子，它不僅可以捕捉彩色影像，更可以捕捉深度資訊，讓各種應用中的3D資訊擷取變得更為方便。 本研究提出了三種方法來達到在3D視訊監控中保護隱私及進行3D影像偽裝的目的。第一種方法可在3D視訊監控中保護使用者所選擇的隱私區域。該法利用可逆的映射函式，將隱私區域的影像轉變為與背景影像相似的偽裝影像，藉以產生一受保護視訊。第二種方法將該法延伸，做到隱藏私密性動態活動的作用，並藉由加速穩健特徵(speeded up robust features, SURF)和匹配演算法(matching algorithm)來自動偵測動態區域，作為私密影像的部分。第三種方法運用可逆的對比映射（reversible contrast mapping, RCM）方法來偽裝3D影像，可以將指定的3D秘密影像轉變成3D偽裝影像，並可無失真地恢復原來3D秘密影像的本貌。根據RCM的方法可以把3D秘密影像隱藏在3D偽裝影像中，藉以產生資訊隱藏的效果。 本論文所提出的方法皆獲得良好的實驗結果，證明其在實際應用上可行。

With the rapid development of stereo-vision technology, 3D imaging devices become more and more popular. Many types of such devices have been invented to acquire 3D multimedia information under various conditions. One famous example is the KINECT device manufactured by Microsoft. It can capture not only color images but also depth images, making it easier to get 3D information for uses in various applications. In this study, we propose three methods for video surveillance applications with the aims of privacy protection in 3D video surveillance as well as 3D image steganography. The first method is proposed for protecting a user-selected privacy-sensitive region in a 3D surveillance video, in which reversible prediction-based mappings are used to convert the privacy-sensitive region in each image frame into a background image part, resulting in a protected video. The method is extended further to conceal private motion activities in surveillance videos, by which motion regions are detected automatically as privacy-sensitive image parts by the SURF extraction and matching algorithm. The reversible mapping scheme used in the two methods guarantees lossless retrieval of the concealed privacy-sensitive image part from each image frame in the protected video. Moreover, a method using the reversible contrast mapping (RCM) scheme for 3D image steganography is proposed, which can hide 3D secret images into 3D cover images, as well as recover the embedded 3D secret image losslessly. Specifically, the method hides 3D secret images into non-object holes in the 3D cover image according to the RCM scheme to yield a 3D stego-image which has a steganographic effect. Also embedded into the input 3D cover image is a recovery sequence which can be extracted from the 3D stego-image to retrieve the original 3D secret image. Good experimental results are also presented to show the feasibility of the proposed methods for real applications.