高藏量極低失真的高動態範圍影像資訊嵌入演算法

High Capacity and Very Low Distortion Data Embedding Algorithms for High Dynamic Range Images

柯衍慶(Yen-Ching Ke) ； 王宗銘(Chung-Ming Wang)

《興大工程學刊》 23卷2期 (2012/07) Pp. 71-89

https://doi.org/10.6287/JENCHU.2012.2302.03

摘要

本文提出三個高動態範圍影像的資訊嵌入演算法，具有高藏量極低失真的特性。第一個演算法簡稱TRICOD（Triplet Coding Algorithm）。我們充分善用所有像素提供的狀態數，故可在不提高影像變動量的前提下，增加秘密訊息嵌入量。相較於最新文獻，TRICOD演算法可提升5.52%～ 5.79%之嵌入量。高動態範圍偽裝影像經色調映射後產生的低動態範圍影像，經量化與視覺化量測並無影像失真。我們所提的第二個演算法簡稱NULPIX（Null Pixel）演算法。此演算法分類高動態範圍影像之像素，將秘密資訊嵌入至像素分類中尚未被使用之像素。相較於最新的文獻，此演算法平均可以提高48.37%之嵌入量，色調映射影像經量化或視覺化之量測下並無影像失真。FEAPIX（Feasible Pixel）演算法是文獻首創的第三個演算法。我們提出一個最佳化計算機制，將秘密資訊嵌入至另一類尚未被使用之像素。FEAPIX演算法可大幅提高2.77～3.02倍之秘密訊息嵌入量。色調映射影像經量化後其PSNR仍高達76.60～84.44dB；影像雖略有些微失真，但人眼視覺化無法辨識差異性。總結本研究，我們提出的三個演算法分別能提供高藏量無失真或具極低失真之特性。演算法均具盲擷取特性且效益優於最新文獻之結果，擴展秘密資訊嵌入在影像註記與藏密學應用之範疇。

關鍵字

高動態範圍 ； 無失真資訊嵌入 ； 極低失真資訊嵌入 ； 影像註記 ； 藏密學

並列摘要

This paper presents three information embedding algorithms for high dynamic range images. Our algorithms offer high embedding capacity where the first and second algorithm provide the embedded image with no distortion and the third algorithm offers very low image distortion. Our first algorithm, TRICOD, makes use of all statuses provided by pixels to increase the capacity that can be concealed in an image. Comparing with the previous work, our algorithm can improve the embedding capacity in the range between 5.52% and 5.79%. No image distortion is encountered when tone mapping the high dynamic range embedded images to produce the low dynamic range embedded image. Our second algorithm is referred to as NULPIX algorithm. It takes advantages of a new pixel category to convey more secret messages. This algorithm can increase the average embedding capacity of 48.37% without causing any image distortion. FEAPIX algorithm is the third method we present. We adopt an optimization computation mechanism to produce a number of potential pixels that produce the least image distortion for information embedding. FEAPIX algorithm can largely increase the amount of embedding capacity with the magnitude between 2.77 and 3.02. The tone mapped image presents high PSNR values (76.60 ~ 84.44 dB) showing no perceivable visual difference. In conclusion, our three algorithms can provide higher capacity than our counterparts and offer the characteristics of no image distortion or extremely low distortion. These algorithms are with blind extraction characteristics, and they are appropriate to information embedding applications such as image annotation and image steganography.

並列關鍵字

high dynamic range images ； distortion-free data embedding ； very low image distortion ； image annotation ； image steganography