Sampling patterns with a blue noise distribution are widely used in many areas of computer graphics, yet their efficient generation remains a difficult problem. We propose a method to generate point sets with a blue noise distribution using a deterministic algorithm with no preprocessing. We insert each new sample at the center of the largest empty circle in the point set, which is obtained by calculating the Delaunay-triangulation of the set and finding the triangle with the largest circumcircle. Our method supports adaptive sampling according to a user-specified density function, as well as specifying the exact number of required samples. It can also be extended to perform sampling on a three-dimensional curved surface.

View full abstract

This paper describes a new approach for passive measurements of transparent glass plate surfaces, in which the position distributions of the surfaces are obtained from virtual images of pattern boards taken with a camera. In the approach, structured patterns displayed on two boards are acquired with a camera through the surface and reflection points on the front and rear surfaces are located geometrically using the triangulation principle without any prior knowledge of the coordinates of other reflection points. This feature enables us to obtain a high spatial resolution of detection with high accuracy. In real-image experiments, reflection point position distributions on the planar glass plates of 3- to 10-mm in thickness were detected. The overall averages of position errors from the true are below 0.2 and 0.5mm along the lateral and vertical directions, respectively. The average of the error absolutes is less than 1mm and the standard deviation is less than 1.12mm for the lateral and vertical directions. Furthermore, it is made clear that the detection errors are caused by the quantization errors. The detection principle and system are simple, and the detection accuracy and spatial resolution, as well as the reliability, seem to be satisfactory enough for use in the inspection and production fields.

View full abstract

This paper describes a new approach for accurately detecting scaling-change center positions of slightly scaling-changed images and also proposes a new technique for correcting the scaling changes. These approaches utilize the phase difference spectrum analysis approach for detecting image shifts and the technique for detecting scaling factors, without iterative approximation calculation. In experiments, three types of image patterns were used. When the position is within a few ten pixels from the image center, the detection accuracy of the scaling-change center position is 2.2 pixels on average and as high as 1.1 pixels, excluding errors for scaling-factor changes below ±0.6%. The correction accuracy for scaling factors is almost completely independent of the scaling-center position and is about 0.0004 on average. The pixel displacement of the corrected image from the reference image is about 0.012 pixels on average when the scaling-change center position is within a few ten pixels from the image center. From these results, it is clear that the proposed approaches, despite their simplicity, make it possible to detect scaling-change center positions for scaling-changed images and to correct the images with high accuracy and reliability.

View full abstract

We propose an MPEG-2 to H.264 intra transcoding for interlace stream of frame and field macroblock intermingled to improve our previous work. This method uses an adaptive MB pair type selection method that keeps as many discrete cosine transform (DCT) coefficients of the original MPEG-2 bitstream as possible to avoid mixture of re-quantization noise. Experimental results show that the proposed method improves the peak signal-to-noise ratio (PSNR) values by about 0.33-1.55dB over those of the conventional methods.

View full abstract

This paper proposes a restoration method for JPEG and blurred images based on ensemble learning algorithm. Such as a median filter, various restoration filters for these images have been already proposed and applied in real images. However, when images have edge and non-edge regions, it is difficult for single filter having fixed coefficients to reduce both blocky and mosquito noises adequately. Here, new restoration methods for JPEG and blurred images are proposed by introducing multiple muliti-layered neural networks. Each neural network can be adapted for edge, flat, and texture regions in images by ensemble learning. From experimental results, the proposed method can obtain good restoration accuracy compared with conventional filters. Moreover, we have confirmed that the proposed method could automatically assign restoration tasks to each neural network according to the characteristics of each region in given distorted images.

View full abstract

In this paper, we propose an image classification method using local features and linear manifolds. In our method, local features such as SIFT are extracted from test and training images. In each class, all local features of training images are compressed by principal component analysis, thus a set of local features extracted from training images is represented by a linear manifold (affine subspace). In a voting phase, dissimilarity between a local feature extracted from a test image and each linear manifold is measured, and one vote is added to the class to which the nearest linear manifold belongs. Through voting by all test local features, the test image is then classified into the class that has majority votes. In this paper, we also adopt a steepest descent based learning algorithm to our classification rule for improving accuracy. The experimental results on the WANG image dataset verifies that our method performs as well or better than a nearest neighbor voting method using all training local features.

View full abstract

This paper describes a detection method of pulmonary nodules (i.e. possible cancers) from chest X-ray computed tomography (CT) scans. The method consists of two phases. First, initial nodule candidates are detected by use of circle- and ring-shaped discriminant filters. Next, false positives such as blood vessels are reduced by use of radially-arranged rectangular discriminant filters and a view-based support vector machine. The proposed method is applied to actual CT scans, and experimental results are shown.

View full abstract

In this paper, linear subspaces are used for detecting and retrieving rare cells called nucleated red blood cells (NRBCs) from microscope images. In our method, NRBCs are detected based on similarities between a candidate region and two subspaces spanned by NRBC images and their similar ones. After that, detected NRBC candidates are displayed to users in ascending order of the dissimilarities defined by the difference of similarities to each subspace. Furthermore, we derive a learning rule for our method for improving detection performance. Experimental results show that our method can detect NRBCs in about one second per microscope image and our learning can reduce the number of false detection.

View full abstract

The digital road map is an important basis for the Intelligent Transport System and Geographic Information System; however, almost all digital maps are currently generated manually from paper maps. As it is very time consuming to generate digital road maps this way, we propose a system to generate digital maps using image processing of aerial images. Conventionally, most methods of road extraction use RGB features. However, the use in urban areas of RGB features has many problems as they are affected severely by shadows cast by buildings. On the other hand, road extraction using a Digital Surface Model (DSM) is not affected by shadows because a DSM is elevation data. However, it is too difficult to distinguish between a road and other objects, such as a railway, because the elevations of these objects are similar. In this paper, we use DSM and RGB aerial images to resolve the problems of each type of data. We segment a DSM to extract the road areas from the DSM; however, the segmentation method conventionally used could not extract narrow roads because that method is sensitive to noise in a DSM. In light of this, we next propose a method of image segmentation using Boundary code. The proposed segmentation method is robust against the noise in a DSM. In an experiment using actual data, the accuracy of road extraction using the proposed method was 83.7%, against the 76.0% achieved using the conventional method. In addition, the proposed method achieves faster segmentation than the conventional method. Moreover, this method extracted road areas more accurately in urban areas than by just using RGB or DSM.

View full abstract

We have investigated a computer simulation method of color perception by elderly people. However, the previous method has two problems to be considered. That is, the luminescence spectrum of the display used in the experiments is not sufficiently examined, and there are some mismatches between the results of the artificial test through this simulator and the ones by the clinical test for elderly people using SPP color perception test. To improve this simulation method totally, in this work, (i) we utilize the actual luminescence emission spectrum measured for the LCD display used in the experiments. Also, (ii) we relax the color constancy hypothesis so that the lenticular yellowing changes not only luminosity but also chromaticity in color perception. Experimental results suggest that the accuracy of the simulator is fairly improved by these modifications.

View full abstract

This paper describes a method for achieving novel appearances of a given face that can create intended impressions on observers. A morphable 3D model of the human face is developed to describe the variations in appearance while also using a small number of parameters. To find the physical parameters of the 3D objects that might strongly contribute to their visual impressions, the impression ratings obtained for learning samples are computationally investigated. Based on the quantification of higher-order impressions by the Semantic Differential (SD) method, the visual impression of the face can be transformed by manipulating the parameters defined in the morphable 3D model. The ability of the proposed method to transform impressions is confirmed by the paired comparison method.

View full abstract

In recent years, many line drawing algorithms have been proposed for 3D shape depiction. One common drawback is that these algorithms draw only part of the necessary lines for fully depicting the shape. Cole et al.¹⁾ have shown that these algorithms can be complementary to each other. Based on that, in this paper, we combine the previous algorithms to produce better line drawings of 3D objects. By directly superposing multiple line drawing algorithms, however, there is a risk of generating too many unnecessary lines, especially in regions where many lines are located near to each other, making them to appear over-sketched. To solve this problem, we define an attribute called “importance” for all points on each line. The importance describes which lines should be drawn in preference. In addition, we determine the width of the lines based on the number of lines in their neighborhood. Our algorithm runs on GPU, achieving real-time speed. From experimental results, our method avoids drawing excessive lines while conveying the shapes effectively.

View full abstract

Creating a 3D model of a human face from an image sequence remains an important problem in the fields of computer vision and computer graphics. We propose a method based on a factorization algorithm; it automatically constructs a 3D face model from an image sequence that contains turntable motion of the face as captured by a monocular camera. Our proposal uses multiple Active Appearance Models for feature tracking to avoid manual operations and to recover the 3D shape of the whole face. In contrast to previous methods, our approach requires no 3D generic model and can obtain the user's individual 3D face shape. Experiments conducted on real image sequences demonstrate the effectiveness of the proposed method.

View full abstract

In this paper, we propose a new type of deform map where both the close and the distant view are expressible. When we use a 2D ordinal map of a local area, we sometimes want to know some information of the outside, i.e. the direction of a far city or the shape of far mountains from the local area. To realize this demand, a new expression is proposed. The local information is drawn in the central area of the map by top view parallel projection like 2D ordinal maps. The information of the far cities and mountains are drawn in the surrounding area by overhead or side view projection so that the direction of cities and the shape of mountains are expressed. To establish smooth connection of both areas, a novel multi aspect projection technique is developed. Several rendering techniques, such as blending of multiple map textures, LOD control, and contour enhancement of mountains, are combined to obtain comprehensible result within interactive time.

View full abstract

In this paper, a method to enable people to experience “Chinkin” in the virtual space is proposed. Chinkin is a traditional artistic technique in Japan and some Asian countries to draw designs on lacquer ware with fine lines and dots, and it was widely used to produce Daimyo's utensils in the Edo period during the 17th and the 19th century. The author develops a virtual Chinkin system based on the virtual sculpting method which is an interactive CG creating method developed by the author. The real Chinkin consists of some processes and each process is realized by virtual carving: an interactive deformation of solids, and virtual painting: an interactive generation of a 3D texture map in the virtual space. Improvement in a rendering method is also given for more useful operation. In this system, the user can experience each process of the Chinkin technique in the virtual space and can create virtual sculptures ornamented with Chinkin designs as CG. Both the operation and the result are similar to the real ones and it would be useful for introduction, education and preservation of the traditional Chinkin technique.

View full abstract

We propose an efficient method based on physical laws to model phenomena near the surface of the sun, which is known as the solar photosphere. In the field of astronomy, some physical models have been developed to simulate the sun's turbulence. Most of the previous models are difficult to be applied in the making of CG images since they require a considerable amount of time even when using a supercomputer. The subject of our research is the visual simulation of the phenomena observed above the photosphere, namely solar prominences, because these are characteristic phenomena of the sun that can greatly influence the visual impact in movies and games. The sun is mainly composed of ionized plasma, and its behavior can be treated as a fluid. However, unlike gas, the plasma fluid is influenced by the magnetic field. Thus, magnetic field calculations are needed to calculate the plasma behavior. We use the magnetohydrodynamics (MHD) equations to simulate the behavior of plasma. We propose a new method which can simulate a prominence in a practical computation time. The computation cost is reduced by simplifying the phenomena inside the sun: we only consider the phenomena after the solar prominence erupts because the phenomena before the eruption dose not manifest itself in a visual way. To render the simulation results, we emulate an observation method that extracts the specific spectrum emission from the solar plasma.

View full abstract

We propose a visual appearance simulation method for exhibited objects such as direction boards considering both eyesight and visual angles of viewers. Our method consists of two steps: optical simulation step and neural simulation step. In optical simulation step, first, the visual angle of the actual exhibited objects and of the images on computer monitors are matched. Then, we generate blurred image to simulate viewers' eyesight by using Gaussian filter. In neural simulation step, we simulate image contrast enhancement due to lateral inhibition by applying a convolution filter with DOG distribution. Appropriate parameters of both filters are based on experimental measurements, which are corresponding to viewers' eyesight. Our method can simulate visual appearance of various exhibited objects which is viewed from arbitrary distance with arbitrary eyesight.

View full abstract

Objective evaluation methods of visual induced motion sickness (VIMS) caused by interactive video such as computer game is studied to clarify the relation between VIMS and image motion components. We focus on time-variant biosignal indices related autonomic nervous activity, and estimate availability of two indices called ΔI_LF/HF based on LF/HF ratio and triggerpoints from heart rate signals. As a result of experiment by 20 subjects, the biosignal indices have positive correlation with the global motion of video in case of non-interactive video, whether unpleasant subjective score is low or high. In case of interactive video, on the other hand, both subjective score and triggerpoints by biosignal decrease compare to that of non-interactive video, and ΔI_LF/HF indices tendency depends on the operation type of video in case of interactive video. These results suggest that triggerpoints indices are possible to use for evaluation of VIMS related by GMV, and ΔI_LF/HF indices are possible to use unconscious VIMS effect in ANA level caused by video motion and operation rather than subjective unpleasant sensation.

View full abstract

We propose the method generating a saliency map using the self-organized filter generted from the observation of the image. So we introduce the ICA base estimation to the model generating a saliency map. Especially, the photo receptor cell is not regular. We extend the ICA base estimation to apply for the non-regular positioned receptor cells. Moreover we generate the model so that a receptor cell can receive the information about color and motion which human receives. We show the effectiveness of the proposed early vision model generating a saliency map based on the ICA base estimation by appling this model for the image.

View full abstract

This paper introduces a new class of CFA interpolation algorithm. Conventional CFA interpolation methods usually apply a directional filtering in horizontal-, vertical-, or omni-direction even if edge runs in diagonal direction. We, then, invent a similarity based filtering using a local map called unified high-frequency (UHF) map. This new technique enables flexible filtering masks so as to be able to deal with edges of any direction such as diagonal edges. It is confirmed through experiments that the proposed algorithm outperforms state-of-the-art methods in terms of PSNR and subjective quality. In addition, the proposed algorithm is comparable to, or less expensive than the state-of-the-art methods regarding the resource consumption.

View full abstract

An optical flow is defined as a motion vector for each pixel between successive two images in a video. This paper proposes a fast optical flow algorithm with robustness for luminance change. The algorithm includes the luminance change estimation, initial vector decision, fast hierarchical processing, and window size optimization. These methods result in the robustness for the luminance change, accurate flow estimation, and low computational costs. Simulation results show the flow accuracy improves from 18.65° to 4.29°, when giving luminance change. The CPU execution time reduces to 15% compared to the conventional with the same accuracy. In addition, we propose a dedicated architecture based on this algorithm and implement it on an FPGA. Experimental results show the new architecture decreases the operating frequency to 20%, the circuit area to 45%, and the power consumption to 9%. This circuit is applicable to real-time image recognition and image reconstruction.

View full abstract

We propose a novel method of feature selection for similar-shaped character recognition using genetic algorithms (GA). The feature is assigned to the chromosome, and value of “1” or “0” is given to the chromosome; corresponding to features that are respectively used and unused for recognition. The proposed method selects only genes in recognition experiments with which the recognition rate of training samples exceeds the predetermined standard as a candidate of the parent gene and adopts a reduction ratio in the number of features used for recognition as the fitness value, and has a mechanism decreasing the number of chromosomes which take the value of “1” while changing generations. This mechanism enables the number of features to be decreased maintaining the recognition rate constantly. On the experiment for similar-shaped character recognition, the proposed method achieved a higher recognition rate and larger decrease of the number of features compared with Fisher's criterion. Moreover, the feature visualization that displays the pixel which is the source of the feature offered an experimental result that GA automatically acquires difference of character shape guessed that man is using for distinction. The effectiveness of the proposed method was confirmed by these.

View full abstract

An ideal voting technology would have four attributes: anonymity, scalability, speed, and integrity. Electronic voting is getting more important and practical due to the progress of computer and internet. In conducting elections, a secured voting system plays a vital role to protect against threats to the integrity of elections. In this paper, we provided a mutual authentication and secured key exchange method by using a onetime code book extracted from the image of the votes. Blind signatures by the election committee and by the voters are embedded in the images according to RSA and MD5 algorithms. Experimental results show the efficiency and the robustness of it.

View full abstract

In this paper Kinoform Type Computer Generated Holographic Stereogram is considered. It uses multitasking operating system and two dimensions Fast Fourier Transform of real 3D object taken of more than one Web-cameras.

View full abstract

Presentation of bilingual messages with Japanese sign language (JSL) and Japanese for deaf people at emergency situations is one of important issues in designing emergency messaging systems. In this paper, we discuss characteristics for eye movements and fixations in reading for three deaf individuals. The results for these analyses show interesting reading behaviours of deaf individuals. When two message areas overlap every subject pays attention on this area. When two message areas are separated, deaf individual pays mainly gaze upon a specific message area. Results for analyses also show variation in message reading behaviour. In the last part of the paper, we point out issues to be considered thoroughly in designing an effective bilingual message system.

View full abstract

[in Japanese]

View full abstract