The huge success of the Internet allows for the transmission, wide distribution, and access of electronic data in an effortless manner. Content providers are faced with the challenge of how to protect their electronic data. This problem has generated a flurry of research activity in the area of digital watermarking of electronic content for copyright protection. The challenge here is to introduce a digital watermark that does not alter the perceived quality of the electronic content, while being extremely robust to attack. For instance, in the case of image data, editing the picture or illegal tampering should not destroy or transform the watermark into another valid signature. Equally important, the watermark should not alter the perceived visual quality of the image. From a signal processing perspective, the two basic requirements for an effective watermarking scheme, robustness and transparency, conflict with each other. We propose two watermarking techniques for digital images that are based on utilizing visual models which have been developed in the context of image compression. Specifically, we propose watermarking schemes where visual models are used to determine image dependent upper bounds on watermark insertion. This allows us to provide the maximum strength transparent watermark which, in turn, is extremely robust to common image processing and editing such as JPEG compression, rescaling, and cropping. We propose perceptually based watermarking schemes in two frameworks: the block-based discrete cosine transform and multiresolution wavelet framework and discuss the merits of each one. Our schemes are shown to provide very good results both in terms of image transparency and robustness.

Image-adaptive watermarking using visual models

This presentation describes how a vision model incorporating contrast sensitivity, contrast masking, and light adaptation is used to design visually optimal quantization matrices for Discrete Cosine Transform image compression. The Discrete Cosine Transform (DCT) underlies several image compression standards (JPEG, MPEG, H.261). The DCT is applied to 8x8 pixel blocks, and the resulting coefficients are quantized by division and rounding. The 8x8 'quantization matrix' of divisors determines the visual quality of the reconstructed image; the design of this matrix is left to the user. Since each DCT coefficient corresponds to a particular spatial frequency in a particular image region, each quantization error consists of a local increment or decrement in a particular frequency. After adjustments for contrast sensitivity, local light adaptation, and local contrast masking, this coefficient error can be converted to a just-noticeable-difference (jnd). The jnd's for different frequencies and image blocks can be pooled to yield a global perceptual error metric. With this metric, we can compute for each image the quantization matrix that minimizes the bit-rate for a given perceptual error, or perceptual error for a given bit-rate. Implementation of this system demonstrates its advantages over existing techniques. A unique feature of this scheme is that the quantization matrix is optimized for each individual image. This is compatible with the JPEG standard, which requires transmission of the quantization matrix.

/pdf/dct-quantization-matrices-visually-optimized-for-individual-2kcsnp29yr.pdf

DCT quantization matrices visually optimized for individual images

The growth of new imaging technologies has created a need for techniques that can be used for copyright protection of digital images and video. One approach for copyright protection is to introduce an invisible signal, known as a digital watermark, into an image or video sequence. In this paper, we describe digital watermarking techniques, known as perceptually based watermarks, that are designed to exploit aspects of the the human visual system in order to provide a transparent (invisible), yet robust watermark. In the most general sense, any watermarking technique that attempts to incorporate an invisible mark into an image is perceptually based. However, in order to provide transparency and robustness to attack, two conflicting requirements from a signal processing perspective, more sophisticated use of perceptual information in the watermarking process is required. We describe watermarking techniques ranging from simple schemes which incorporate common-sense rules in using perceptual information in the watermarking process, to more elaborate schemes which adapt to local image characteristics based on more formal perceptual models. This review is not meant to be exhaustive; its aim is to provide the reader with an understanding of how the techniques have been evolving as the requirements and applications become better defined.

/pdf/perceptual-watermarks-for-digital-images-and-video-1qmg8lxfta.pdf

Perceptual watermarks for digital images and video

The discrete wavelet transform (DWT) decomposes an image into bands that vary in spatial frequency and orientation. It is widely used for image compression, measures of the visibility of DWT quantization errors are required to achieve optimal compression. Uniform quantization of a single band of coefficients results in an artifact that we call DWT uniform quantization noise; it is the sum of a lattice of random amplitude basis functions of the corresponding DWT synthesis filter. We measured visual detection thresholds for samples of DWT uniform quantization noise in Y, Cb, and Cr color channels. The spatial frequency of a wavelet is r2/sup -/spl lambda//, where r is the display visual resolution in pixels/degree, and /spl lambda/ is the wavelet level. Thresholds increase rapidly with wavelet spatial frequency. Thresholds also increase from Y to Cr to Cb, and with orientation from lowpass to horizontal/vertical to diagonal. We construct a mathematical model for DWT noise detection thresholds that is a function of level, orientation, and display visual resolution. This allows calculation of a "perceptually lossless" quantization matrix for which all errors are in theory below the visual threshold. The model may also be used as the basis for adaptive quantization schemes.

Visibility of wavelet quantization noise

https://faculty.washington.edu/jpalmer/palmer_verghese_pavel00e.pdf

The psychophysics of visual search.

A detection model is developed to predict visibility thresholds for discrete cosine transform coefficient quantization error, based on the luminance and chrominance of the error. The model is an extension of a previously proposed luminance-based model, and is based on new experimental data. In addition to the luminance-only predictions of the previous model, the new model predicts the detectability of quantization error in color space directions in which chrominance error plays a major role. This more complete model allows DCT coefficient quantization matrices to be designed for display conditions other than those of the experimental measurements: other display luminances, other veiling luminances, other spatial frequencies (different pixel sizes, viewing distances, and aspect ratios), and other color directions.

Improved detection model for DCT coefficient quantization

Studies of visual detection of a signal superimposed on one of two identical backgrounds show performance degradation when the background has high contrast and is similar in spatial frequency and/or orientation to the signal. To account for this finding, models include a contrast gain control mechanism that pools activity across spatial frequency, orientation and space to inhibit (divisively) the response of the receptor sensitive to the signal. In tasks in which the observer has to detect a known signal added to one of M different backgrounds grounds due to added visual noise, the main sources of degradation are the stochastic noise in the image and the suboptimal visual processing. We investigate how these two sources of degradation (contrast gain control and variations in the background) interact in a task in which the signal is embedded in one of M locations in a complex spatially varying background (structured background). We use backgrounds extracted from patient digital medical images. To isolate effects of the fixed deterministic background (the contrast gain control) from the effects of the background variations, we conduct detection experiments with three different background conditions: (1) uniform background, (2) a repeated sample of structured background, and (3) different samples of structured background. Results show that human visual detection degrades from the uniform background condition to the repeated background condition and degrades even further in the different backgrounds condition. These results suggest that both the contrast gain control mechanism and the background random variations degrade human performance in detection of a signal in a complex, spatially varying background. A filter model and added white noise are used to generate estimates of sampling efficiencies, an equivalent internal noise, an equivalent contrast-gain-control-induced noise, and an equivalent noise due to the variations in the structured background.

Visual signal detection in structured backgrounds. II. Effects of contrast gain control, background variations, and white noise

Several recent image compression standards rely upon the discrete cosine transform (DCT). Models of DCT basis function visibility can be used to design quantization matrices for arbitrary viewing conditions and images. Here we report new results on the effects of viewing distance and contrast masking on basis function visibility. We measured contrast detection thresholds for DCT basis functions at viewing distances yielding 16, 32, and 64 pixels/degree. Our detection model has been elaborated to incorporate the observed effects. We have also measured detection thresholds for individual basis functions when superimposed upon another basis function of the same or a different frequency. We find considerable masking between nearby DCT frequencies. A model for these masking effects also is presented.

Discrete cosine transform (DCT) basis function visibility: effects of viewing distance and contrast masking

Object detection involves looking for one of a large set of object subimages in a large set of background images. Image discrimination models predict the probability that an observer will detect a difference between two images. We find that discrimination models can predict the relative detectability of objects in different images, suggesting that these simpler models may be useful in some object detection applications. Six images of a vehicle in an otherwise natural setting were altered to remove the vehicle and mixed with the original image in various proportions. Nineteen observers rated the 24 images for the presence of a vehicle. The pattern of observer detectabilities for the different images was predicted by three discrimination models. A Cortex transform discrimination model, a contrast sensitivity function filter model, and a root-mean-square difference predictor based on the digital image values gave prediction errors of 15%, 49%, and 46%, respectively. Two observers given the same images repeatedly to make the task a discrimination task rated the images similarly, but had detectabilities a factor of two higher.

Models of human image discrimination predict object detection in natural backgrounds

Object detection involves looking for one of a large set of object sub-images in a large set of background images. Image discrimination models only predict the probability that an observer will detect a difference between two images. In a recent study based on only six different images, we found that discrimination models can predict the relative detectability of objects in those images, suggesting that these simpler models may be useful in some object detection applications. Here we replicate this result using a new, larger set of images. Fifteen images of a vehicle in an other-wise natural setting were altered to remove the vehicle and mixed with the original image in a proportion chosen to make the target neither perfectly recognizable nor unrecognizable. The target was also rotated about a vertical axis through its center and mixed with the background. Sixteen observers rated these 30 target images and the 15 background-only images for the presence of a vehicle. The likelihoods of the observer responses were computed from a Thurstone scaling model with the assumption that the detectabilities are proportional to the predictions of an image discrimination model. Three image discrimination models were used: a cortex transform model, a single channel model with a contrast sensitivity function filter, and the Root-Mean-Square (RMS) difference of the digital target and background-only images. As in the previous study, the cortex transform model performed best; the RMS difference predictor was second best; and last, but still a reasonable predictor, was the single channel model. Image discrimination models can predict the relative detectabilities of objects in natural backgrounds.

Andrew B. Watson

Papers

Improved detection model for DCT coefficient quantization

Visual signal detection in structured backgrounds. II. Effects of contrast gain control, background variations, and white noise

Discrete cosine transform (DCT) basis function visibility: effects of viewing distance and contrast masking

Models of human image discrimination predict object detection in natural backgrounds

Image Discrimination Models Predict Object Detection in Natural Backgrounds