English  |  正體中文  |  简体中文  |  Items with full text/Total items : 43312/67235
Visitors : 2106436      Online Users : 6
RC Version 5.0 © Powered By DSPACE, MIT. Enhanced by NTU/NCHU Library IR team.

Please use this identifier to cite or link to this item: http://nchuir.lib.nchu.edu.tw/handle/309270000/154558

標題: 可預測混合式的可回復資訊嵌入演算法之研究
A Study of Predictable and Hybrid Reversible Data Embedding Algorithms
作者: 郭慧彬
Guo, Huei-Bin
Contributors: 王宗銘
Chung-Ming Wang
資訊網路多媒體研究所
關鍵字: 可回復式資訊隱藏;直方圖位移法;預測機制;混合預估;高動態範圍影像;色調映射;視覺差異評估
reversible data hiding;histogram shifting;prediction mechanism;hybrid predictor;high dynamic range images;tone mapping;visual difference predictor
日期: 2012
Issue Date: 2013-11-21 10:56:27 (UTC+8)
Publisher: 資訊網路多媒體研究所
摘要: 可回復式資訊隱藏-直方圖位移法是一種適用於醫學、軍事影像的資訊藏密之應用技術。藉由可回復的特性,使得嵌入訊息後的偽裝影像仍能夠完整的回復成原始影像。本論文提出兩個使用直方圖位移法、適用於灰階影像與高動態範圍影像的可回復式資訊隱藏演算法。
我們提出的第一個演算法為具預測機制的可回復式資訊隱藏演算法。該演算法算法適用於灰階影像。藉由此預測機制,我們能評估未嵌入訊息前的影像之相關的數據結果。在預測機制的步驟上,我們必須先考慮秘密訊息“ 0 ”與“ 1 ”的分佈機率情形,並且評估掩護影像的像素值來建立對應的五個直方圖。接著,我們根據此預測機制所提出的預測運算來計算出嵌入效率、嵌入量以及影像品質。最後,便可依據使用者的需求來提供相對應的數據結果。此預測機制也可以擴展到對影像資料庫上的預測,這使得我們可以更快速的從影像資料庫中找到最適合使用者所需求的影像。其中,我們捨棄以單一像素預估器來預估像素值,於本篇論文首次提出混合六種像素預估器來提高預估像素值的準確度,並且進一步地來增加訊息的嵌入藏量。實驗結果顯示,我們可以藉由此預測機制,即可評估出掩護影像嵌入訊息後的各種數據。而且,我們根據使用者依嵌入量或影像品質所提出的條件,也能預測出滿足使用者條件之結果。以四個影像資料庫而言,我們使用此預測機制更能迅速地找到最佳嵌入效率之影像。其中,我們混合六種預估器對掩護影像的嵌入量做出評估,所評估出的準確率可以高達99.9%。實驗結果也顯示,使用混合的六種預估器對四個公用資料庫可以提供5.75%~32.15%不等之額外嵌入量。
我們提出的第二個演算法為混合式高動態範圍影像可回復資訊嵌入演算法。此可回復式演算法利用混合式嵌入技術來提高嵌入量,演算法可依照使用者給予的限制條件適用在高動態範圍影像偽裝學與影像註記兩種應用。演算法共分三個步驟。首先利用六個預測器來預測像素值,根據預估誤差產生直方圖後,求出該層級最佳預估器。接著,依直方圖位移方法嵌入秘密訊息並作多層級嵌入。我們提出的嵌入技術可以根據使用者所輸入的不同應用之限制條件,來產生出相對應用的結果數據。對於輸入的一張RGBE格式高動態範圍影像,我們的演算法維持E通道上的數值,僅對R、G、B三通道進行混合式秘密訊息嵌入。我們以色調映射的技術來將高動態範圍影像轉換成低動態範圍影像,並藉由視覺差異記算的方式來評估出兩張影像在訊息嵌入前後之差異程度。評估後,我們即可根據產生的數據,依照給定的限制條件,來決定是否在進行下一層級的訊息嵌入動作。影像偽裝學應用時,我們以影像品質與偽裝偵測為考量;影像註記應用則以高動態範圍影像所能嵌入的最大訊息量為基準。相較於使用單一預估器,使用我們提出的混合式預估器對偽裝學應用平均可以提供4.47%~14.39%的額外訊息嵌入量,對於影像註記可以提供高達10.05%~16.63%之額外訊息嵌入量。嵌入前後的影像相關係數都能維持在0.9以上的高度相關。
本論文之貢獻為提出兩個利用直方圖位移的可回復式資訊隱藏演算法,分別適用於灰階影像與高動態範圍影像。第一個演算法能有效的預測出實際嵌入訊息後的結果;而第二個演算法能提供偽裝學與影像註記應用。實驗結果證實此兩演算法成效優異、具體可行,藉由擴展以直方圖位移法達成可回復資訊影藏之實際應用領域。
Reversible data hiding using the histogram shifting method provides a desirable characteristic to fully recover the original cover image after the message extraction. This paper proposes two reversible data hiding algorithms based on histogram shifting method for grayscale images and high dynamic range images.
The first scheme we provide is a prediction mechanism for reversible data hiding. Given a cover image, we construct and analyze five corresponding histograms. We further operate the prediction computing taking into account the probability distribution of the secret bit “0”and “1.”Consequently, our scheme can report the embedding efficiency, embedding capacity, and image quality prior to the real message embedding. We extend our scheme to an image database. This allows us to suggest the most appropriate image achieving the highest efficiency within an image database. During the pixel prediction, a variety of six predictors is employed. This greatly improves the prediction accuracy and increases the embedding capacity that can conceal in a cover image. The experiments show that the proposed prediction scheme can foresee the embedding information without practically embedding secret messages into a cover image. The scheme can recommend a proper image satisfying the user’s demands for the image quality and the embedding capacity. The approach of using the hybrid predictor leads to a faithful prediction where the prediction accuracy in average can be as high as 99.9% within four test image databases.
The second scheme we present is a hybrid reversible data embedding algorithm for high dynamic range (HDR) images. The histogram shift technique is employed to embed messages and achieve the reversibility. Given an HDR image at each embedding plane, our hybrid approach adopts six predictors to derive the corresponding histograms of prediction errors, based on which we can determine one of predictors that performs with the highest embedding efficiency as the best predictor. Our scheme is suitable for steganographic application and image annotation, depending on the given constraints. With regard to the first application, a multiple plane embedding cannot be proceeded unless the visual difference between the cover and stego images is still under the constraints derived by the VDP and VDP2 visual difference evaluations. In the image annotation, however, the constraint is devoted to maximizing the pure capacity that an HDR image can offer by taking into consideration the magnitude of the recorded overhead information needed to achieve the reversibility. Experimental results show that our hybrid predictor approach outperforms a single predictor method. It provides an average of 4.47%~14.39% extra capacities for steganographic application and introduces an average of 10.05% ~ 16.63% additional capacities for image annotation application. The resultant low dynamic range stego images produced by tone mapping the HDR stego images show high image qualities with the PSNR values over 30 dB. The Pearson correlation coefficients between 12 pairs of cover and stego images are around 0.9 indicating images are highly correlated even conducting the message embedding. To the best of our knowledge, our hybrid algorithm is the first in the literature capable of embedding messages and achieving reversibility for high dynamic range images.
In conclusion, our proposed algorithms offer six significant features: (1) the prediction mechanism can accurately predict the results prior to message embedding; (2) the scheme can response to user’s demanding for the image quality and the embedding capacity; (3) the most suitable image within an image database can be recommended which can achieve the best embedding performance; (4) the hybrid predictor approach can supply extra embedding capacity than a single predictor does; (5) based on different constraints, the scheme is flexible to apply to steganography as well as image annotation applications; (6) both the correlation and the visual difference evaluations show little image difference is encountered between the cover and stego images.
Appears in Collections:[依資料類型分類] 碩博士論文

Files in This Item:

File SizeFormat
index.html0KbHTML137View/Open


 


學術資源

著作權聲明

本網站為收錄中興大學學術著作及學術產出,已積極向著作權人取得全文授權,並盡力防止侵害著作權人之權益。如仍發現本網站之數位內容有侵害著作權人權益情事者,請權利人通知本網站維護人員,將盡速為您處理。

本網站之數位內容為國立中興大學所收錄之機構典藏,無償提供學術研究與公眾教育等公益性使用。

聯絡網站維護人員:wyhuang@nchu.edu.tw,04-22840290 # 412。

DSpace Software Copyright © 2002-2004  MIT &  Hewlett-Packard  /   Enhanced by   NTU/NCHU Library IR team Copyright ©   - Feedback