Cross-sectioning is one of the most commonly used methods for visualizing the complicated inner structures of three-dimensional volume datasets. However, the method usually requires a tedious trial-and-error process by users and thus often fails in finding characteristic cross-sections that capture the underlying inner structures. This paper presents a method for automatically generating characteristic cross-sections from a given volume dataset. The application of a graph structure called volume skeleton tree (VST) allows us to automatically extract such characteristic cross-sections by delineating the topological structure of the given volume. The feasibility of the proposed method is demonstrated using simulated and measured volume datasets.

View full abstract

In this paper, we propose a method for directly rendering point sets which only have positional information by using recent graphics processors (GPUs). Almost all the algorithms in our method are processed on GPU@. Our point-based rendering algorithms apply an image buffer which has lower-resolution image than a frame buffer. Normal vectors are computed and various types of noises are reduced on such an image buffer. Our approach then produces high-quality images even for noisy point clouds especially acquired by 3D scanning devices. Our approach also uses splats in the actual rendering process. However, the number of points to be rendered in our method is in general less than the number of input points due to the use of selected points on an image buffer, which allows our approach to be processed faster than the previous approaches of GPU-based point rendering.

View full abstract

We present a real-time rendering method of translucent materials such as marble for large 3D models. To render translucent objects, it is necessary to take into account the subsurface scattering of light. However, rendering translucent objects with the subsurface scattering takes long time. In our method, virtual points are set around the object uniformly. The intensities of the subsurface scattering of light are calculated at each virtual point. The intensities of points that represent the translucent object are interpolated by the intensities of the virtual points. Our method can render about 2,000,000 points with the subsurface scattering in real-time.

View full abstract

A color edge detection method is proposed for content-based image retrieval by sketches. In content-based image retrieval by sketches using edge-based image features, it is essential to detect edges of a target object. In this method, images are first classified into principal five kinds of colors for chromatic color and two kinds of colors for achromatic color, based on Munsell color system, and then edges are detected for each classified colors. Experimental results of applying to 5,500 images in Corel Photo Gallery show that this method is effective in retrieving images with unrelated background to target and with plural objects.

View full abstract

Image segmentation can be viewed as a pixel classification problem, where one would assign a label to each pixel within an image in order to specify the class to which the pixel should belong. As with the estimation of labels, it is not easy to determine optimal values for model parameters. Our model proposed in this work could be effective in cases of the lack of sufficient information by introducing to the priors of parameters a hierarchical structure that would make minimal assumptions on the data. Markov chains in Gibbs sampling would efficiently explore the joint posterior of segmentation labels and model parameters. We can use samples obtained by simulating this chain to perform Bayesian inference of labels. We attempt a quantitative evaluation of image segmentation algorithms by comparing their results to those given by humans.

View full abstract

In the ubiquitous society, text information is one of the most important communication methods and a variety of text information are always displayed on cellular phone, navigation system, DTV, etc. Therefore, it is strongly demanded to improve the legibility of text on displays. However, typical design process of fonts for specific devices requires a long development period, whereas print type has been gradually developed in long history. In this paper, considering the gravity center of characters, which makes a large impact to the legibility, evaluation method of gravity center based on potential energy have been proposed. The experimental results show that the potential energy-based gravity center on characters can reflect the position of strokes in silhouette of character and improve the computation accuracy of gravity center on characters.

View full abstract

In the field of intelligent transport systems or security systems, development of the image processing method which tracks moving objects on the image taken by a fixed camera is a very important subject. In this paper, we propose new tracking method, considering the renewal of background by the time passage, which has not been done in the former methods. Then, block division of the radial reach information in the pixel unit which is calculated by Radial Reach Filter, known as one of the typical methods to detect the object, is carried out, and it applies to the distinction of a moving object. Through the experiments using the real image, this paper shows the effectiveness of our proposed approach.

View full abstract

3D cinemas are becoming real thanks to digital image processing technology. The most feasible and stable technology based on the binocular disparity requires stereo-glasses. However, there is still a strong desire for glasses-free method in order to ease the stress to viewers. In this paper, Space-Sampling method is investigated for glasses-free 3D movies. With a layered image sensor device, Space-Sampling method captures 3D image projected through a lens. Layered display panels reproduce 3D images. They can be projected on a refractive screen to reconstruct the original 3D image space. Transparent image sensors, layered display panels, depth resolution, blur elimination and occlusion problems are discussed.

View full abstract

This paper describes a real-time vision system that recognizes the location and position of automobile cylinder sleeves on a moving conveyor belt based on surface texture pattern. The proposed vision system will maximize production efficiency and improve the strenuous and labor-intensive conditions of assembly lines. The normalized correlation function uses multi-resolution pyramid images to quickly find candidates. To discriminate these candidates, the output scores of the normalized correlation function are classified and the higher order autocorrelation functions are employed. The results from field tests prove that the performance is fully capable in providing a basis for a practical vision system. In the future, by integrating other surface texture patterns, the proposed system can expand the combined real-time vision system for the factory automation field.

View full abstract

To identify a 3D coordinate of an object from photographic images, corresponding point of each image must be known. Though it is possible to obtain corresponding points by the simple pattern matching among similar images from near viewpoints, it is difficult among images from arbitrary viewpoints. In this paper, we examined the matching method which adds the corresponding points automatically for multiple view images. In this method, template of each image is decided according to an affine transformation by referring three known corresponding points, and a new corresponding point is obtained by pattern matching along the epipolar line. As a result, it was confirmed that correct corresponding points are obtained on some images of the object with plane or curved surface.

View full abstract