Tadahiko Kimoto

Implementation of level transformations for hiding watermarks in image bit-planes under limited level changes

In hiding watermarks in bit-planes of a grayscale image, this paper focuses on reduction of level change caused by bit inversion for improving image quality. First, the level transformation to invert one of bits representing a pixel level while minimizing the whole change is defined. According to its level change properties, a scheme for implementing the transformations so that actual level changes can appear below given limits is presented. In addition, domains of the transformation are defined so that watermarks can be recovered from the transformed images alone. On the other hand, allowable level changes are estimated according to human visual properties. Simulations using test images demonstrated effects of the level transformation on preserving image quality. Also, an example of limited level changes was given and then, an implementation scheme for achieving the limitation was specified.

Modified Level Transformation for Bit Inversion Inwatermarking

The transformation function proposed in a previous paper performs both inverting a bit chosen in a multi-level signal and minimizing the resultant change of level. The remaining level changes, however, get larger in the intervals including the ends of the source level range than in the middle intervals. The principal object of the present paper is to remove this difference in transformation performance among the intervals and thus to obtain the same properties over all the source range. To avoid the ends of the level range, a translated and extended coordinate system is proposed. The transformation is then modified and formulated so as to involve the translating of levels. Furthermore, a function of varying the output levels in a random manner is added to the modified transformation. Consequently, the output levels can spread over the whole dynamic range. The effects of the random variations are evaluated through a stochastic analysis.

An Advanced Method for Watermarking Digital Signals in Bit-Plane Structure

Inverting signal bits is a basic operation for data hiding such as digital watermarking. In the previous papers the level transformation that performs both inverting a specified bit and minimizing the resultant level change has been proposed. Because the transformation maps signal levels sparsely within the dynamic range, some variation of the transformed levels is necessary to conceal the fact that the signals are transformed. This paper proposes a method for varying the transformed levels randomly so that those bits that present the randomness can be distinguished. Then, these bits can represent a message to recover. Accordingly, a capacity of embedding additional bits increases. The performance of the proposed method is analyzed in a stochastic manner in terms of the resultant level distortion and an embedding capacity. A simulation result is also shown to demonstrate the actual performance in the application to digital images.

Measurement of subjective quality of watermarked images made by inverting bits

Inverting signal bits is a basic operation for digital signal processing such as watermarking. A transformation from an input level to an output level that performs all the following three functions has been proposed in the previous papers: first, inverting a chosen bit; second, minimizing level change caused by the inversion; and last, varying the resultant transformed levels randomly by a stochastic process. In the application of the transformation to watermarking image bit-planes this paper deals with distortions appearing on the transformed images in detail. First, the way how pixel levels are changed by the transformation to generate distortions is described. Second, the results of the subjective quality measurement that was carried out for human subjects are presented. The measurement indicates that subjective quality of low-detail areas shows a correlation with level change, and subjective quality of high-detail areas reveals a correlation with change of frequencies of level occurrence.

Error-Correction Coding of Spatial Image Watermarks in Gray Code

This paper deals with robustness of a spatial-domain-based digital watermark against signal distortions caused by lossy data compression usually used in digital communication systems, and presents a method for encoding a watermark bit in the Gray codeword representing a spatial signal by a (3, 1, 3) single-error-correcting (SEC) code. The two code words in the SEC code corresponding to respective watermark bits are determined so as to minimize the expected amount of embedding distortions. The stochastic analyses also show a capacity of the SEC code for changes in signal code words. Experiments using 8-bit monochrome images were carried out to evaluate the performance of the proposed method, where the watermarked images were distorted by the JPEG DCT-baseline coding scheme. The results demonstrate that, compared with a conventional method, the proposed method yields watermarked images with less distortions and achieves the comparable robustness.

Implementation of modified level transformation for bit-plane watermarking

The inverting of signal bits is a basic operation for signal processing such as image watermarking. For performing the bit inversion for pixels, a level transformation that performs all the following three functions has been proposed in previous papers: first, inverting the bit associated with the chosen bit; second, reducing the change in the whole level caused by the inversion; and last, varying the transformed levels by a stochastic process. Procedures to implement the transformation are presented in this paper. Amounts of level change and changes of frequency distribution of levels caused by the transformation are also considered through simulations using real image data. The simulation results demonstrate that the stochastic properties of the transformation for real images can be approximately estimated from the properties obtained by an analysis under the assumption of a uniform frequency distribution of source levels.

Evaluation of Effectiveness of Random Variations in Signal Levels Transformed for Data Hiding

In a method of hiding data in bit-planes that correspond to a series of digital signals, a single bit is chosen from the bits of each processed signal and then inverted. In a previous work a level transformation that performs the following three functions has been proposed: inverting the chosen bit, minimizing the level change caused by the bit inversion, and yielding random variations in the output level. This paper evaluates the random variations resulting from the transformation and demonstrates its effectiveness in secure data hiding. First, the properties of the level transformation are analyzed. Some theoretical values are also calculated especially to evaluate the increase of level differences caused by the random variations. Next, results of computer simulations of the transformation are presented, and the changes in level distributions are measured. These results indicate that the random variations in output levels reduce effectively the deformation of the level distributions at the expense of a little increase of the level differences

A Perceptually Adaptive Scheme for Image Bit-Inversion-Based Watermarks

A perceptually adaptive scheme for image watermarking by bit inverting transformation is developed. The transformation performs the change of a signal level by both inverting a bit and changing the level at random in a limited range. From the property analysis of the transformation, a perceptual model of two metrics of level distortion is assumed for evaluating a subjective visual quality of the transformed image. The measurements of image quality by the human subjective evaluation are analyzed and then, the evaluation function that can yield an estimate of the subjective visual quality from the two distortion measurements of the transformed image is determined. By using the evaluation function, every image region is transformed with its parameter adaptively changed so as to achieve a desired visual quality. The simulation results demonstrate the performance of the evaluation function and show a linear correlation between the evaluated values and the measurements of visual quality.

A subjective image quality metric for bit-inversion-based watermarking

An image watermarking scheme using the previously proposed bit embedding method is developed. To achieve a desired subjective visual quality in the watermarked image, the embedding parameter that is related to both the image quality and the embedding capacity is determined by using a perceptual model. First, based on the properties of the bit embedding method, the perceptual model of two kinds of objective quality measures is assumed. Then, the measurements of human subjective image quality are analyzed from the viewpoint of the correlation with these two measures. Thereby, the estimating function that can yield an estimate of the subjective quality from two objective measurements is determined. According to the estimating function, the bit embedding method performs in each image region so as to achieve a desired subjective image quality while increasing the capacity of embedding watermark bits. The simulation results demonstrate that the estimating function values have a linear correlation with human subjective evaluations, and the embedding parameters can be adaptively changed in every image region by using the function.

Image Segmentation of Pastel Drawings

What is examined in drawing therapy is the picture contents of a clients drawing. Aiming at computer-aided drawing therapy, this paper deals with an image processing technique for extracting the whole region which represents the contents from a pastel drawing. A handwritten pastel drawing often has a distinguishing feature of coarse textures, which make the image segmentation by simple thresholding difficult. Also, the picture contents of a drawing can be usually interpreted in various ways. Considering these points leads to a method for determining the whole region of picture contents in a pastel drawing image so that the requirements of the practice of drawing therapy can be met. In the method, first, painted pixels are obtained by thresholding. Then, the set of the pixels are fused by morphological processing, and thereby the region boundary is closed. The method has variables in each processing so that the resulting region can be varied. Experimental results demonstrate how effectively each variable performs in varying the figure of a resulting region according to the interpretation of picture contents.