Lossless Data Hiding For Digital Images

Throughout history, great changes have taken place in the way of humancommunications from language, writing to digital multimedia. The past years havewitnessed an explosion in the availability of digital images in different fields due tosuch factors as direct viewing, vividness and super-language. However, the fastdevelopment of the information technology has brought to forefront security concernswith the use of digital images. Recently, data hiding has been proposed and popularlyemployed as an effective way to solve this problem. Usually, traditional data hidingmethods cause irreversible distortion of image contents during the embedding process,which is an obstacle to their usage in such important applications as medical diagnosis,remote sensing and law enforcement. Therefore, lossless data hiding (LDH) of digitalimages has attracted increasing attention for copyright protection in multimediainformation security.To target the challenging problems of LDH, the paper focuses on reversibility,imperceptibility, embedding model and robustness of LDH and makes an extensivestudy of lossless data hiding for digital images. The main contents of this paper aresummarized as follows.(1) An improved reversibility LDH method is proposed based on the optimalstrategy. First, the stable regions for watermark embedding are picked up by the optimalstrategy. Secondly, a novel embedding mechanism together with a parameter model isdesigned to tune the embedding process. Based on this, the proposed method caneffectively overcome the unstable reversibility of the histogram distribution constraintmethod.(2) By taking the importance of imperceptibility to data hiding into consideration, acontent-adaptive reliable LDH method is designed on the basis of the histogramrotation-based embedding model. The host images are first preprocessed by the pixeladjustment strategy to avoid both the overflow and underflow problems of pixels, andthus to improve the visual quality of watermarked images. Thereafter, inspired by theperceptual characteristics of the human visual system, the watermark strength isoptimized with the help of the luminance masking model, leading to thecontent-adaptive, reliable and lossless data hiding.(3) Due to the diversity of grayscale histograms of different images, watermark embedding suffers from poor stability and adaptability. Therefore, we develop a novelframework for LDH using statistical characteristics and histogram-based neighborhoodselection. The generalized statistical quantity histogram (GSQH) is first employed todeduce the diversity of grayscale histograms. Based on this, the stable and tunableembedding regions are constructed by combining the GSQH and histogram-basedneighborhood selection, leading to flexible capacity control as well as improvement ofstability and adaptability. Moreover, the encryption and lossless compression techniquesare utilized to solve the problem of side information storage, which is helpful toenhance security of the proposed method.(4) In practical applications, watermarked images may be degraded byunintentional attacks to some extent. Aiming at such an issue, we propose a robustlossless data hiding (RLDH) method, which incorporates the merits of GSQH andclustering algorithms. Based on the investigation of the distribution of embeddingregions in watermarked images, the k-means clustering algorithm is adopted to dealwith their dynamic division to improve the robustness. Extensive experimental resultshave demonstrated the superiority of the proposed method from the aspects ofrobustness, imperceptibility, capacity, adaptability and stability.(5) To further improve robustness against unintentional attacks, a novel RLDHmethod based on visual perceptual models in the wavelet domain is proposed, whichemploys the statistical characteristics of wavelet coefficients in high-pass sub-bands.The property inspired pixel adjustment is first utilized to solve both the overflow andunderflow problems of pixels. Following this, watermark embedding and extraction isachieved by incorporating the histogram shifting and clustering techniques. Furthermore,we design an enhanced pixel-wise masking model to optimize watermark strength,leading to a good tradeoff between imperceptibility and robustness.In this paper, we integrate the basic theories of pattern classification and computervision into LDH and propose five novel LDH methods. By effectively overcoming thedrawbacks of the existing methods including unstable stability, poor visual quality androbustness, the proposed ones will be able to successfully address the challengingproblems facing the LDH.