[in Japanese]

View full abstract

Research has continued on the shooting conditions for production of natural stereoscopic images. In this paper, we present an analysis of the conversion of images from real space into stereoscopic images from the point of view of the geometrical relationships of the binocular parallax. In particular, the differences in conversion between images shot by a parallel camera configuration and those shot by a toed-in camera configuration are discussed, taking into consideration the effects of the horizontal shift between left and right images. We also derived some simple equations relating to the distortions peculiar to stereoscopic images, such as the puppet theater effect and the cardboard effect. According to these equations, stereoscopic images shot using a parallel camera configuration do not produce the puppet theater effect while a toed-in camera configuration may produce the effect. Both camera configurations may produce the cardboard efffect.

View full abstract

We have been researching and developing an image space that based mainly on a photo-realistic texture, that can be shared by people in different locations and that can lend a real sense of being present. This paper proposes a technique for creating a setting representation of a natural panorama. The setting representation is a compact pseudo 3D expression using approximate representation on planes, which enables virtual views to be generated. The scene is captured using an original three-camera system. The images are stitched into stereo panoramas and the depth is estimated. The (texture) panorama is segmented into regions, each of which can be regarded approximately as a plane. The set of plane parameter for each region is estimated depending on the depth data (setting representation). Based on the setting representation the virtual views are calculated using the center panorama, and the left and right panoramas are used for occlusion compensation.

View full abstract

3D video is the ultimate medium for recording images of a dynamic visual event in the real world. The recorded behaviors of objects can be observed from any viewpoint, because 3D video records whole 3D shape, motion, and precise surface properties (i.e. color and texture) of objects.In our last paper, we presented a method of reconstructing the shape of a dynamic 3D object from multi-view video images, which enables a temporal series of 3D voxel representations of an object's behavior to be obtained in real-time. In this paper, we present (1) an algorithm for generating texture on the surface of a 3D object from multi-view video images, and (2) an editing system for visualizing 3D video with an omnidirectional background image using a versatile 3D camera action. This paper focuses on how we can generate high fidelity object images from arbitrary viewpoints based on 3D object shapes of limited accuracy. We propose a novel texture-mapping algorithm, which maps textures onto the surface of 3D object depending on the specific viewpoint. Experimental results demonstrate, the effectiveness of the algorithm in generating high fidelity images of objects from arbitrary viewpoints.

View full abstract

We have developed a system called the Stereo Omnidirectional System (SOS). The stereo units, which are composed of three cameras, in this system are arranged on each plane of regular icosahedrons. This system can acquire all directional color and stereo pair images, in real time. However calibrating this multi-camera system, which acquires images in different directions from the center of the system, is difficult because simultaneously acquiring images of common regions from multiple cameras is not possible. We described a hybrid calibration method in which the translation parameters use the system design parameters and the rotational parameters are obtained from the images. These adjustments are made because of the difficulty in accurately, obtaining the translation and rotational parameters from the images so that they can be compared. We also described how the color registration method uses common region images acquired from the system.

View full abstract

To achieve realistic 3D displays using by computer-generated holograms, both modeling and rendering methods are key techniques. In this paper, we propose a new algorithm for object-light calculation considering reflectance distribution, which is based on fast processing of spatial frequency. The method is applied to Phong's reflection model, which is a well-known model in computer graphics. The results of optical experiments show that the method can be successfully applied to computer-generated holograms.

View full abstract

In the field of computer graphics and 3-D visual communication, ray-based representation of 3-D scenes/objects has been studied. The most popular way to acquire light ray data is to take perspective projection images from multiple viewpoints, however, light rays are equivalent to orthogonal projections. This difference in projection causes errors in images synthesized from light rays. To solve this problem, we have developed a capturing system using telecentric optics, which provides orthogonal projections of objects. This system captures light rays with an angular resolution of 0.09° by rotating object (s) and taking orthogonal projection images automatically. In this paper, we discuss the implementation of the telecentric capturing system. In particular, we focus on the errors caused by disarrangement of the system's components or lens distortion, and show some methods for measuring and correcting errors. Furthermore, experimental results show the applicability of the system for acquiring light ray data as orthogonal projections and synthesizing arbitrary views.

View full abstract

We propose a new method for modeling complex range data using a parametric model. We first compute local surface curvatures by quadratic fitting of a local surfaces, and then extract the curved surfaces and plane regions from the range data by examining the probability distribution of the local surface curvatures. After extracting these regions, one shape can be divided into multiple regions using the characteristics of the feature space. Then, after the parameters of each region have been estimated, the fitting error between the regions of the integration candidates are calculated. Next, by applying statistical testing techniques to the fitting error, we establish the integration criteria and integrate the regions that satisfy the criteria. Using this process, it is possible to perform a robust surface extraction. Experimental results on real range data show the effectiveness of the proposed method.

View full abstract

We propose an algorithm for automatic simplification of models of 3D objects based on a simplification technique that applies vertex-clustering. The major advantages of the vertex-clustering technique are its low computational cost and high data reduction rate. Thus, the method is suitable for use in real-time rendering.
This paper presents a clustering method for better preservation of important features in the shape. Its major contributions include : improved tight octree, described by Luebke, and high accuracy as a result of estimating vertices for the distance between the octants at each level.

View full abstract

[in Japanese]

View full abstract

The Kanade-Lucas (KL) feature tracker is one of the better matching methods. However it fails to find corresponding points of two images when image motion is quite large. We propose the feature point matching method, making use of affine transformation and correlation so that the KL method works well for large image motion. First, the KL method is used for matching points. Some points are matched well, other points are mismatched or failed to be matched. Second, correlation of two matched points between the first and the second image obtained by the KL method is used to discriminate between well matched points and badly matched points. If the correlation coefficient is lower than a specified threshold, it is regarded as an outlier. Third, all the points to be matched in the first image are transformed on the second image by affine transformation of well matched points alone. However, all the transformed points from the first image to the second image are not well matched well due to the movement of the image. The role of affine transformation is to reduce the computation time and the number of mismatching points by designating only a small searching area around the transformed point. Finally, correlation is used to find the best matching point around the transformed point. Real image data has been used to test the proposed method, and excellent results have been obtained with an average error of 0.811 pixels.

View full abstract

In this paper, we propose a new feature for image retrieval using a rough sketch. The proposed feature is based on edge shape and is used to represent the relative position of the edge pixels. This edge-based feature is shift-, scale and rotation-invariant. In this method, the number of edge pixels in each region of the pixel of interest is divided radially into eight directions and is then counted and thresholded to represent the pixel by an 8-bit binary pattern. The resulting histogram, containing 256 bins corresponding to 8-bit words in counter-clockwise order, is regarded as an image feature. To verify the effectiveness of this method, we applied the method to two image databases : 1, 650 images from musical instruments and 3, 200 images from the COREL photo gallery. The results showed that the proposed feature is effective for image retrieval.

View full abstract

[in Japanese]

View full abstract

The receiving antenna polarization that maximizes SINR (Signal to Interference and Noise Ratio) was determined in the presence of varying arbitrarily polarized interferences and thermal noise using the genetic algorithm (GA) and Poincaré sphere representation of the polarization state. SINR is maximized by controlling a receiving antenna polarization that is mismatched to both the desired signal and interference polarizations so that a balance between the interference and the signal is obtained. The optimal receiving antenna polarization in the presence of varying interference and thermal noise was calculated using the GA. The relationship between the desired signal, the interference, and the optimal receiving antenna polarizations is shown. These results provide insight into applying polarization theory to improving signal reception in the presence of interference and thermal noise.

View full abstract

A 1/3-inch 640×480-pixel CMOS image sensor was developed using a simple fixed-pattern-noise-reduction technology with a five-transistor pixel circuit and a low input-voltage I-V converter. In this report, we show the effectiveness of a low input-voltage I-V converter with a current-mirror circuit in improving the amplification ratio and linearity of a pixel circuit. The dependence of the pixel signal characteristics on the parameters of the pixel transistors was also analyzed. In a five-transistor pixel circuit, the threshold voltage of the X-Y addressing transistor affects the amplitude and level of the readout pulse. This report also contains analysis of the mechanism of the X-Y addressing transistor, illustrating the concept behind the selection of the threshold voltage.

View full abstract

[in Japanese]

View full abstract