[in Japanese]

View full abstract

[in Japanese]

View full abstract

We studied the synergism of resolution and gray-scale levels in image recognition in relation to the precise phase detection in vernier acuity. Humans who have a 1.0 visual acuity candetect 1.0 minute-size (1/60 degree) objects. Video systems are designed based on this data. On the other hand, it has been reported that human can detect an approximately 2.0-second (2/60/60 degree) phase shifts. We found that precise position detection is possible with a low resolution detector if the number of gray-scale levels is sufficient. We applied this principle to the position adjustment on a multi-screen display system.

View full abstract

In the case of very low bit-rate video coding such as MPEG-4 for small display devices, the encoding parameters of the image size and the frame rate must be set appropriately. Although a subjective quality evaluation is necessary for judging the image quality, a method for evaluating small image display systems has not been developed yet. We propose a new subjective method for evaluating the quality of such systems, and we study the encoding parameters. First, we extract two psychological factors : “image quality” and “continuity”. Next, we verify that the overall evaluation can be expressed by a linear combination of evaluation scores of the two extracted psychological factors, and that the results of the overall evaluation provide useful guidelines for setting the encoding system parameters. We show that the results for QSIF size tend to be better than those for SIF size for the same bit-rate.

View full abstract

This paper discusses a video-telecommunication method that can transmit real-time video with high quality and economical path. In an information recovery technique, this paper proposes a method based on a new approach to motion vector interpolation. We also propose a new recovery method using an associative memory model, for macroblocks that cannot use motion vectors. Simulation results showed that this method keeps video quality degradation to a minimum on poor quality telecommunication paths.

View full abstract

The distortion due to low frequency cutoff degrades the performance of partial response maximum likelihood (PRML) system in a perpendicular magnetic recording. This paper presents an application of guided scrambling (GS) code to PRML system as an effective strategy for avoiding the distortion. The bit error rate performances of PR2ML systems for two types of GS codes, Galois field (GF) additive type and convolutional operation type, are obtained. The results indicate that GS coded PR2ML systems can achieve gains of about 6.3dB for GF additive type code and 4.4dB for convolutional operation type code over conventional Reed-Solomon coded PR2ML system at a normalized linear density of 1.5, a power ratio of jitter-like media noise to total noise of 80% and a low-cutoff frequency normalized by bit rate of 0.002.

View full abstract

The acquisition of the 3D shape of real objects is a rapidly expanding field with various applications in Cornputer Aided Design, archeology, design, entertainment and so on. In this paper wepropose a new approach for 3D object reconstruction from range data. Our approach is based on the modelization of the object surfaces with a set of RBFs (Radial Basis Functions). Our originality is the fact that the reconstruction is independent from the object geometry; it can be applied for smooth objects as well as for objects with sharp boundaries. To enhance the efficiency of the RBF modelization, we propose also an efficient isosurfacing algorithm. Unlike conventional reconstruction techniques, no prior information about the acquisition process is required. We demonstrate the ability of our approach to reconstruct a wide range of 3D objects through a set of experiments.

View full abstract

With recent advances of technology, mobile phones are generally used as convenient communication tools. One of desired functions for the mobile phone is a nonverbal communication system thatwould display facial expressions. This function is expected to deliver more intimate communication via the phone. To create the system, effective extraction of a face, hair and individual facial features is one of the most important issues. This paper proposes a Mahalanobis-distance-based method for analyzing and extracting those features from a facial image to draw a facial portrait. Our extraction experiment is performed using face images of 15 examinees. The method's accuracy of extracting facial features is quantitatively evaluated.

View full abstract

We describe a tiled-projection display system whose entire image is composed of twenty four patches of projection, sixteen of which are used for front projection and eight for rear projection on the center area of the screen. We use eight pairs of projectors to project the stereoscopic images, the left-eye image, and the right-eye image. The screen consists of differently shaped surface elements that are the partial surfaces of objects such as planes, spheres, cylinders, and tori. The following three technologies were developed for the display system : (1) a digital camera based measurement of the position and color features of the pixels projected on the screen, (2) software to compute the geometry correction and color modulation for edge blending by using the measurement data, and (3) real time image processing hardware to enable arbitrary geometrical warping and color modulation for individual pixels.

View full abstract

The induction field in vision means the potential field that represents the effect on the periphery from the source of the visual stimulus. We measured this field as follows. We presented a small spot of light in the neighborhood of the induction figure. The brightness of the light when the stimulus could not be seen was recorded, while the brightness was changed. We created an experiment environment where this procedure could be carried out with a PC-based system. We evaluated the conditions necessary to measure the induction field in vision with this system. Up to now, many researchers have believed that only the contours of an induction figure affect the induction field in vision. This means there are no changes in the induction field, even if the figure is filled, and even if the contour width is changed. We found that the induction field in vision changes, when the figure was filled or the contour width was changed.

View full abstract

We developed a space diversity system for relay sytations in digital terrestrial broadcasting that has a transversal adaptive filter for interference equalization and a coordinating adaptive synchronizer in each branch. By use of equal gain combining or imroved combining, in a computer simulation, the system functions without estimating the frequency characterisitics of transmission paths. We also showed that the diversity operates effectively even in environment with long delay multipath or loop coupling.

View full abstract

Distortion minimization oriented bit assignment technology is representative of approaches that have been studied to improve MPEG-2 rate control performance. This technology conducts an inclusive bit assignment for each picture while maintaining picture distortion at a minimum level. Its effectiveness has been tested and confirmed. In this paper, MPEG-2 HDTV encoding was assumed, and the distortion minimization oriented bit assignment technology was introduced as a basic scheme. We proposed additional improvement approaches with full use of the technology for improving macroblock quantization control and adaptive picture type selection. Coding experiments confirmed that an additional coding gain could be achieved with the proposed scheme, especially for low bit-rate encoding.

View full abstract

In mixed reality that merges real and virtual worlds, it is required to interactively manipulate the illumination conditions in a virtualized space. In general, specular reflections in a scene make it difficult to interactively manipulate the illumination conditions. Our goal is to provide an opportunity to simulate the original scene, including diffuse and specular relfections, with novel viewpoints and illumination conditions. Thus, we propose a new method for estimating diffuse and specular reflectance parameters using the geometries and images of real scenes. The experiment results showseveral estimates of surface reflectance properties and the relighting of scenes under novel viewpoints and illumination conditions.

View full abstract

[in Japanese]

View full abstract

[in Japanese]

View full abstract

[in Japanese]

View full abstract