透過您的圖書館登入
IP:18.217.228.35
  • 期刊

使用調整預估之高動態範圍影像可回復式資訊隱藏演算法

A Reversible Data Hiding Algorithm for High Dynamic Range Images Using the Adjustment Prediction

摘要


本文提出一個使用調整預估之高動態範圍影像可回復式資訊隱藏演算法。我們使用OpenEXR的高動態範圍影像格式為掩護影像,並運用四個預估器與配合我們提出的調整預估法,來計算預估誤差。接著,我們將此預估誤差來嵌入秘密訊息,藉此達成掩護影像可回復性的目標。我們的演算法僅將秘密訊息嵌入像素的尾數,並不更動像素的符號位元與指數。我們提出調整預估法,並將此法其他兩個預估法做比較;第一個為整數預估法。該法將原始像素的尾數視為整數,使用預估器計算出預估像素值;將掩護像素尾數減去此值來產生預估誤差,然後根據此差值來嵌入秘密訊息。第二個為浮點數預估法;該法將掩護像素浮點數值,使用預估器計算出預估浮點數像素值。接著,求出預估誤差後即可嵌入秘密訊息。最後,我們所提的調整預估法則是將掩護像素浮點數值;其次,使用預估器,產生預估浮點數像素值;接著,比較掩護像素指數與預估像素指數之大小;最後,根據比較之結果調整預估像素值,再產生預估誤差來嵌入秘密訊息。實驗結果顯示,調整預估法比其他兩個預估法相比,可產生較小的預估誤差,在相同門檻值下,可以容許更多的像素嵌入訊息,故能提供最高的嵌入量。高動態範圍偽裝影像經過色調映射處理產生低動態範圍偽裝影像,其影像品質仍可維持在30dB以上。HDR-VDP視覺化差異評估顯示掩護與偽裝影像被人眼察覺有差異之機率極低,且掩護與偽裝影像之直方圖具有高度的相關性。總結本文,就我們所知,我們提出的演算法是文獻首創可以適用在OpenEXR格式;調整預估法可提供最高的嵌入量;四個預估器中以MED預估器最為準確;掩護與偽裝影像之直方圖具有高度相關性且視覺差異極低。我們認為演算法可以擴展可回復式資訊隱藏在高動態範圍影像之應用。

並列摘要


In this paper, we propose a reversible data hiding algorithm for high dynamic range (HDR) images using the adjustment prediction scheme. The cover images of interest are HDR images encoded by the OpenEXR format. Given a predictor, our adjustment prediction scheme computes the prediction errors before adopting the difference expansion algorithm to embed secret messages and achieve the reversibility. Our algorithm embeds secret messages into the mantissa field of the pixel making the sign and the exponent fields intact. We compare our algorithm with other two prediction schemes including the integer prediction scheme and the floating-point prediction scheme. Experimental results show that the adjustment prediction scheme can produce the smallest prediction errors, thus providing the highest embedding capacity. Applying the tone mapping operation to an HDR stego image will produce a low dynamic range stego image with a good image quality, where the PSNR values are still over 30 dB. The visual difference analysis using the HDR-VDP has shown that there is a significantly low probability for human eyes to detect any visual difference between the cover and stego images. Finally, cover and stego images produce high correlation coefficients. In conclusion, the proposed adjustment prediction scheme offers the highest embedding capacity; stego images are not likely to be detected with any visual difference; cover and stego images are highly correlated. Our algorithm is feasible to extend the reversible data hiding applications for high dynamic range images.

延伸閱讀