This paper presents two histogram preserving JPEG steganographic methods aiming at secure JPEG steganography against histogram-based attacks. The first one is a histogram quasi-preserving method, which uses quantization index modulation (QIM) at quantization step of DCT coefficients. Since a straightforward application of QIM causes a significant histogram change, a device is introduced in order not to change the after-embedding histogram excessively. The second one is a histogram preserving method based on histogram matching using two quantizers with a dead zone. In comparison with F5 as a representative JPEG steganography, the two methods show high performance with regard to embedding rate, PSNR of stego image, and particularly histogram preservation.

This paper discusses an attack to BPCS-Steganography (Bit-Plane Complexity Segmentation-Steganography) and presents a countermeasure. BPCS is an image-based steganographic method. BPCS embeds secret data by replacing blocks that appear noise-like on bit-planes. Blocks on bit-planes are categorized as a "noise-like region" or an "informative region" by means of the binary-image feature called complexity. When the complexity distribution of noise-like blocks is different from the complexity distribution of the secret data that replaces those blocks, we can see an unusual shape in the form of a valley in the complexity histogram that represents the relative frequency of occurrence of the various complexities. This would be a signature of BPCS. Because steganography must hide the existence of the secret message, the signature should be removed from steganographic images. We present a method for making embedded binary patterns that quasipreserves the complexity distribution.

This paper presents a steganography method using lossy compressed video which provides a natural way to send a large amount of secret data. The proposed method is based on wavelet compression for video data and bit-plane complexity segmentation (BPCS) steganography. In wavelet-based video compression methods such as 3-D set partitioning in hierarchical trees (SPIHT) algorithm and motion-JPEG2000, wavelet coefficients in discrete wavelet transformed video are quantized into a bit-plane structure and therefore BPCS steganography can be applied in the wavelet domain. 3-D SPIHT-BPCS steganography and motion-JPEG2000-BPCS steganography are presented and tested, which are the integration of 3-D SPIHT video coding and BPCS steganography and that of motion-JPEG2000 and BPCS, respectively. Experimental results show that 3-D SPIHT-BPCS is superior to motion-JPEG2000-BPCS with regard to embedding performance.

We have already proposed a large capacity steganography for gray scale images called BPCS-steganography. This paper shows a method to apply BPCS-steganography to palette-based images which consists of a palette storing color vector information and an index image whose pixel value is corresponding to a index in the palette. A palette-based images can be represented by combining R G and B color component images. We embed secret information into the G images. A number of color vectors in a palette after embedding by BPCS would be over the maximum number. In order to reduce the number of colors, the rest of the two component images are then changed in a way that minimizes the square error. The idea behind the color quantization is that the degrading of images manipulated to reduce color is worse than the degrading which occurs with the embedding.

This paper presents a steganography method based on a lossy wavelet compression scheme and bit-plane complexity segmentation (BPCS) steganography. This method utilizes the embedded zerotree wavelet (EZW) compression scheme, where wavelet coefficients of an image are quantized into a bit-plane structure. The proposed steganography enables us to use lossy compressed images as dummy files in bit-plane-based steganographic algorithms. Large embedding rates of around 25% of the compressed image size were achieved with little noticeable degradation in image quality. The proposed method can be applied to other wavelet-based lossy compression schemes like SPIHT and JPEG2000, because in these compression schemes the wavelet coefficients are also quantized into a bit-plane structure.

This letter presents a steganography method based on a JPEG2000 lossy compression scheme and bit-plane complexity segmentation (BPCS) steganography. It overcomes the lack of robustness of bit-plane-based steganography methods with respect to lossy compression of a dummy image: a critical shortcoming that has hampered deployment in a practical scenario. The proposed method is based on a seamless integration of the two schemes without compromising their desirable features and makes feasible the deployment of the merits of a BPCS steganography technique in a practical scenario where images are compressed before being transmitted over the network. Embedding rates of around 15% of the compressed image size were achieved for preembedding 1.0-bpp compressed images with no noticeable degradation in image quality.

One difficulty of textured image segmentation in the past was the lack of computationally efficient models which can capture statistical regularities of textures over large distances. Recently, to overcome this difficulty, Bayesian approaches capitalizing on computational efficiency of multiscale representations have received attention. Most Previous research has been based on multiscale stochastic models which use the Gaussian pyramid decomposition as image decomposition scheme. In this paper, motivated by nonredundant directional selectivity and highly discriminative nature of the wavelet representation, we present an unsupervised textured image segmentation algorithm which is based on a multiscale stochastic modeling over the wavelet decomposition of image. For the sake of computational efficiency, versions of the EM algorithm and MAP estimate, which are based on the mean-field decomposition of a posteriori probability, are used for estimating model parameters and the segmented image, respectively.

In speaker recognition there are usually a small number of speakers whose utterances are difficult to be recognized correctly and then recognition errors are mainly come from those speakers. Recognition performance for those speakers may be improved if they are urged to produce more utterances. Such speaker-dependent utterance-length control has recently been realized in speaker verification (SV) using the sequential probability ratio test (SPRT). The SPRT is in principle for two-class classification and therefore it was naturally applied to SV. This paper implements the speaker-dependent utterance-length control using the SPRT first in speaker identification (SI), by making it applicable to multiclass classification. Experimental results show that the proposed SI method is superior on computation time as well as error rate, to a conventional method with fixed-length utterances.

This paper proposes a Markov random field (MRF) model-based method for unsupervised segmentation of multispectral images consisting of multiple textures. To model such textured images, a hierarchical MRF is used with two layers, the first layer representing an unobservable region image and the second layer representing multiple textures which cover each region. This method uses the expectation-maximization (EM) method for model parameter estimation, where in order to overcome the well-noticed computational problem in the expectation step, we approximate the Baum function using mean-field-based decomposition of a posteriori probability. Given provisionally estimated parameters at each iteration in the EM method, a provisional segmentation is carried out using local a posteriori probability of each pixel's region label, which is derived by mean-field-based decomposition of a posteriori probability of the whole region image.

This paper is concerned about speaker veri ca tion SV using the sequential probability ratio test SPRT In the SPRT input samples are usually as sumed to be i i d samples from a probability density function because an on line probability computation is required Feature vectors used in speech processing obviously do not satisfy the assumption and there fore the correlation between successive feature vectors has not been considered in conventional SV using the SPRT The correlation can be modeled by the hidden Markov model HMM but unfortunately the HMM can not be directly applied to the SPRT because of statistical dependence of input samples This paper proposes a method of HMM probability computation using the mean eld approximation to resolve this problem where the probability of whole input samples is nominally represented as the product of probability of each sample as if input samples were independent each other

This paper describes a new technique to embed secret data into a dummy image by using image segmentation based on a local complexity measure. The key idea to this approach is that a binary image can be categorized as "informative" and "noise-like" regions which are segmented by a "complexity measure". If the embedding data is noise-like, we can hide it in the noise-like region of the dummy image. If a part of embedding data is simple, then we apply "image conjugate" operation to it. This operation transform a simple pattern into a complex pattern. In our experiment, we could embed two color images in a 512/spl times/512 (8 bits/pixel) size gray image (which was dummy) without losing any information. The total amount of the two embedded images was 115 KB, which was about 45% of the dummy image.

Hilbert curve is one of the space-filling curves published by Peano. There are several applications using this curve such as image processing, computer hologram, etc. In this paper, we concentrate on a lossy compression technique for a gray image using the Hilbert curve. The merit of this curve is to pass through all points in a quadrant, and it always moves to the neighbor quadrant. Our method is based on this neighborhood property, by a simple segmentation of the scanned one-dimensional data using a zero order interpolation. From our experiments, we have confirmed that in spite of the simple computation in comparison to JPEG, acceptable quality images can be obtained at bit-rates above 0.6 bit/pixel.

This paper proposes a Markov random field (MRF) model-based method for unsupervised segmentation of images consisting of multiple textures. This method uses a hierarchical MRF with two layers, the first layer representing an unobservable region image and the second layer representing multiple textures which cover each region. This method is an iterative method based on the framework of the expectation and maximization (EM) method. We make use of an approximation for the Baum function in the expectation step. This reduces the parameter estimation to the maximum likelihood (ML) estimation given the current estimate of the region image. An estimation of the region image (image segmentation) is carried out by a deterministic relaxation method proposed by us.

A data compression technique using a bit-plane decomposition strategy of multivalued images is described. Although the bit-plane decomposition is mainly used for image transmission, our method takes the image expression for image database into consideration. It has two merits which are a hierarchical representation using depth-first (DF) expression and a simple noise reduction algorithm for the DF expression that is similar to human perception. The DF expression is useful for image expansion, rotation, etc. We study the information in an image that should be eliminated by noise reduction. Noise-like patterns in an image are uniformalized and the edge and smooth surfaces remain nearly unchanged. They are not blurred, but instead are a little enhanced. We also study the properties of the black-and-white (B/W) boundary points on bit-planes. The algorithm of the uniformalization process with a DF-expression of an image is described. An experiment for real image data is carried out by a comparison to other methods, and the results are discussed.>

There have been many new developments in interactive analysis for multi-spectral images in the research of remote sensing. In general, the methods used are linear transformations such as principal component analysis. In this paper, the authors present a new interactive method for classifying multi-spectral images using a Hilbert curve which is a one-to-one mapping and preserves the neighborhood as much as possible. This method is based on a hierarchical histogram expression with different resolutions for the mapped one-dimensional data. The classification on this expression can be performed easily instead of using N-dimensional data directly. In order to realize the real time response from the system, the authors make use of data tables storing the addresses and the occurrences of data, etc. Here the address is defined by using the coordinates in N-dimensional space, and is made use of dealing with a part of mapping which can not preserve the neighborhood. In the experiments using LANDSAT image data, it is confirmed that the user can get the real time response from the system after once making the data tables.

The classification of remotely sensed multi-spectral data using classical statistical methods has been worked on for several decades. There have been many new developments in neural network (NN) research, and many new applications have been studied. It is well known that NN approaches have the ability to classify without assuming a distribution. The authors previously proposed an NN model to combine the spectral and spatial information of LANDSAT TM images. In this paper, the authors apply the NN approach with a normalization method to classify multi-temporal LANDSAT TM images in order to investigate the robustness of their approach. From the authors' experiments, they confirm that their approach is more effective for the classification of multi-temporal data than the original NN approach and maximum likelihood approach.

SD-Forms are syntactically categorized as concept label, prescribed form, connected form, statement form and emotion form. Guidelines to SD-Form usage are described by showing examples. The SD-Form works as an interlingua of natural languages. It is also useful for knowledge description of a system. As an application, the authors built an English and Japanese conversational textbase having SD-Forms as meaning data. The source material was a series of conversational English study-textbooks for a radio program in Japan. The textbase was used in an English and Japanese conversational-text retrieval system.

There have been many new developments in neural network (NN) research, and many new applications have been studied. The classification of remotely sensed multispectral data using classical statistical methods has been worked on for several decades. Among the multispectral data, we concentrate on the Landsat-5 Thematic Mapper (TM) image data which has been available since 1984. Using this classical maximum likelihood approach, a category is modeled as a multivariate normal distribution; however, the distribution for Landsat images is unknown. It is well known that NN approaches have the ability to classify without assuming a distribution. We apply the NN approach to the classification of Landsat TM images in order to investigate the robustness of this approach for multi-temporal data classification. The authors confirmed that the NN approach is effective for the classification even if the test data is taken at the different time.>

A method for determining the position of a camera using four point-targets is studied. three rotation angles and a translation vector are used to describe the position of the camera for a pinhole model. For solving the six unknown parameters, a minimum of six point-targets is required to define the matrix uniquely (rotation and translation). However, it is shown that by using the properties of the matrix this number can be reduced to four. The error properties of this method are discussed using real image data.>

The depth-first picture expression (DF-Expression) is an effective data compression technique for binary images. When one applies it to real equipment, one needs a practical encoding/decoding scheme which should be fast and simple in the operation. The authors study two types of coding algorithm, one is the top-down and the other is the bottom-up methods. These algorithms are put into two C programs to compare the speed. The authors have designed an encoding/decoding hardware device, and built it on a breadboard which is linked to a workstaton. The hardware encoder/decoder can achieve a 50-100 times faster operation than the C programs run on a 4 MIPS workstation.>

Discrete space-filling curves are not uniquely defined. In addition to the condition that the curve must pass all the points of the array only once, continuously, it is necessary to add some critera to select the best curves. The authors aim is to preserve two-dimensional continuity as much as possible. The weighted sum of the distances of the points in the curve is minimized, where the weights are inversely proportional to the spatial distance between the points. However, the minimum is not unique. Particularly, space-filling curves always come on symmetric pairs. The generation of a near optimal space-filling curve is done hierarchically.>

A bit-plane data-reduction scheme based on some characteristics of human image perception is described. This scheme is on information lossy/modifying type. General properties of bit-plane-decomposed images are reviewed. A method that modifies less important portions on the bit-planes into flat areas, to achieve more uniform images, is introduced. A hierarchical algorithm for the purpose is presented as the actual strategy. The important features of this algorithm are that it can use the coded data in the depth first expression and that the data modification is not likely to affect edges and isolated small areas, which are both very important for image perception.>

A novel image thresholding method called minimal complexity measure thresholding (MCMT) is presented. It is based on the idea that image thresholding should produce the simplest picture possible, which in turn is based on the idea that a simple pattern is easier for humans to perceive. A hierarchical MCMT (HMCMT) algorithm is also shown; it categorizes portions of each local image area as 'significant' and 'less significant'. The image thresholding operation can be performed by a set of logical operations on the image bit-planes. This operation is called the logical convolution on the bit-planes (LCBP). The DFE (depth-first picture expression) is useful for LCBP operation, because the coded bit-planes are directly processed to produce the threshold pattern in the DFE.>

The depth-first picture expression (DF-expression) is a compact quadtree expression providing a high data compression capability. It is primarily a hierarchically structured data representation for binary pictures. The authors first review the basic ideas of the DF-expression, and its potential data-compression competence. Then the practical problem is studied of how much reduction can be achieved data by converting the expression to a binary coding. It is also shown that there are two directions in the DF-expression strategy for the gray-image coding. One is the bit-plane based, and the other is the three-dimensional (three-valued) DF-expressions. For most of the natural images, the bit-plane based strategy is more effective than the three-dimensional one. The reason for that is discussed.>

A picture coding strategy titled DF-expression (Depth-First picture expression) is now studied from another point of view. The basic idea of DF-expression is briefly reviewed at first. Its capability in data compression is demonstrated using 1024 × 1024 binary pictures. Then the new aspects of DF-expression are studied in reference to picture processing algorithm on the coded form. They include circular shiftings, spectrum of primitives, logical operations, etc. Application of DF-expression to gray images (or multivalued pictures) is the next topics of the paper. DF-expression is applicable to bit-plane coding of any binary image data. Our conclusion with this point is that the gray code is the best binary code system for DF-expression in information preserving sense. Information-lossy type data reduction is another topics in this paper. The authors propose a new data reduction technique in terms of uniformalization of excessively complicated regions in bit-planes. Experimental study follows using a 256 × 256 sized fourbit test picture. Finally, conclusions and other possibility of the applications are remarked.

A method of representing a binary pictorial pattern is developed. Its original idea comes from a sequence of terminal symbols of a context-free grammar. It is a promising technique of data compression for ordinary binary-valued pictures such as texts, documents, charts, etc. Fundamental notions like complexity, primitives, simplifications, and other items about binary-valued pictures are introduced at the beginning. A simple context-free grammar G is also introduced. It is shown that every binary-valued picture is interpretable as a terminal sequence of that G. The DF-expression is defined as the reduced terminal sequence of G. It represents the original picture in every detail and contains no surplus data for reproducing it. A quantitative discussion about the total data of a DF-expression leads to the conclusion that any binary-valued picture with complexity less than 0.47 is expressed by the DF-expression with fewer data than the original ones. The coding algorithm of original data into the DF-expression is developed. It is very simple and recursively executable. Experiments were carried out using a PDS (photo digitizing system), where test pictures were texts, charts, diagrams, etc. with 20 cm × 20 cm size. Data compression techniques in facsimile were also simulated on the same test pictures. Throughout these studies it was made clear that the DF-expression is a very effective technique as a data compression for binary pictorial patterns not only because it yields high data compression but also because its coding and decoding algorithms are very feasible.