[in Japanese]

View full abstract

New large displays are replacing the traditional Cathode Ray Tube. An important goal is to improve the quality of TV images to suit these displays. To achieve this goal, we use an interpolation process that convolutes the TV signal. In particular, a C-type fluency sampling function reported by produces high-quality interpolated images. However, the convolution processes with the C-type kernel requires large-scale hardware and a long delay. Therefore, we need develop a simple hardware implementation method for the interpolation based on the C-type function convolution of the TV signal. A hardware implementation method for the interpolation that can reduce the hardware and timing costs is described. The results that our method can reduce such costs and improve the image quality.

View full abstract

A novel lossless coding scheme for video signals is described. The scheme uses a block-adaptive 3D prediction method that predicts a video signal based on both the current and motion-compensated previous frames. The resulting prediction errors are encoded by using a kind of context-adaptive arithmetic coder. Several coding parameters such as prediction coefficients and motion vectors are iteratively optimized for each frame so that the overall coding rate can be a minimum. Moreover, a quadtree-based variable block-size motion compensation technique is introduced for efficient representation of motion information. Experimental results show that the coding rates of our scheme are 11-17% lower than those of the H. 264-based lossless coding scheme.

We apologize for reversing the pages 1052 and 1053 of this article (pp. 1051-1058). The correct version will be published in the August issue.

View full abstract

A high-efficiency video coding method using matching pursuits, which is a waveform coding technique, was proposed. This method encodes a motion compensated prediction error image by applying matching pursuits. The matching pursuits algorithm approximates a signal using a dictionary. Therefore, the performance of this algorithm depends heavily on the dictionary. We propose a matching pursuits coding method that encodes the sub-band images derived from motion compensated prediction error images. The dictionaries used in our method are constructed to be suitable for the frequency characterisitc of each sub-band image. The complexity of our method is reduced compared with the full-band matching pursuits because of lighter inner product computation due to a reduction in both the resolution of the sub-band image and the basis function length of the dictionary. We evaluated the coding performance and computational complexity of our method via computer simulations.

View full abstract

A method for generating all-in-focus images by integrating multiple differently focused images is described. First, based on the assumption that objects are defocused by a simple geometrical blurring model, we derive a formula between acquired images on certain imaging planes and spatial information of objects by using a convolution of a 3-D blur. Then, we analyze spatial frequency components of the blur kernel to obtain a 3-D filter that generates an all-in-focus image from the acquired images. Simulations of image generation are performed utilizing synthetic images to study how noise, object depths, and certain parameters of the blur and the filter affect the quality. We also introduce pre-processing that corrects the size of acquired images and a simple method for estimating the parameter of the 3-D blur. Finally, we show experimental results of all-in-focus image generation from real images.

View full abstract

Conventional image display devices and image acquisition devices consist of regularly located pixels. The pixels are located in a matrix for ease of implementation. Matrix placement of pixels intrinsically has directional singularity in the representation of images. The clarity of represented images is significantly dependent on the directions that objects in the image, such as lines, are facing. For example, horizontal lines are perfectly represented by matrix pixels, while the slanted lines have the jagged edges.
We developed a pseudorandom pixel placement architecture that has no directional singularity in the representation of images, and we evaluated its characteristics and layout implementation.

View full abstract

We developed a two-scan algorithm for discrete Voronoi tessellations of digital images. The computation time of our method is independent of the number of Voronoi sites. In addition to the previous additively, multiplicatively, compoundly weighted Voronoi diagrams and the additively weighted power Voronoi diagram, we proposed a new weighted Voronoi diagram, namely the compoundly weighted power Voronoi diagram. These five weighted Voronoi diagrams were efficiently computed with our two-scan algorithm.

View full abstract

Multi-directional face image acquisition of moving people and techniques for recognizing them are described. Our system uses an extra active camera in addition to a conventional fixed camera. When the fixed camera detects a person who moves toward the active camera, it zooms in towards the person's head and captures close-up multi-directional face images by using multi-directional-averaged-face templates. It captures 3.9 times as many multi-directional face images as the technique using the front averaged-face template only. Our technique substantially increases the chance of identifying a person in a looking-at-people system. Further more, the obtained face image is recognized by using the template matching technique. The recognition system showed a correct recognition rate of 86.0 % for the face images automatically obtained from people in experimental environments. Our technique would be useful for this system to recover people's names if face-recognition fails at a specific place such as a room entrance, a person steps out of view and quickly comes back into view, or an original ID is reinstated after occlusion ends.

View full abstract

We studied a precise positioning system for a spinstand with a linear actuator and a digital controller. The linear actuator was composed of an air bearing and a linear voice coil motor. The digital controller was consisted of a digital signal processor. We tested a reference response and an impulse disturbance rejection for the linear actuator by using only a feedback controller. However, when we applied only the feedback controller for the linear actuator, there was a tradeoff between the reference response and the impulse disturbance for its performance. We applied a two-degrees-of-freedom (2DF) control which was consisted of a feedback controller and a feedforward controller. The feedback controller suppressed the disturbance and the feedforward controller improved the performance of the reference response. We investigated the effect of the controller from the simulations and the experimental results.

View full abstract

We propose an image-based rendering method called “layered probability mapping.” The algorithm requires modest computer power but produces high-quality output images even from textureless or noisy input images. Thus, real-time applications, such as immersive teleconferencing and live broadcasting, are promising areas of use. The key idea is a layered probability map representation using a set of two-dimensional probabilistic functions. This takes into account ambiguities regarding object depth. To test the feasibility of this method, we developed a prototype system with nine synchronized cameras and a thirteen-PC cluster. By calculating the algorithm in parallel, we obtained stable 5-15FPS image generation without any off-line procedures between the image capture and the display.

View full abstract

A novel operation scheme that creates combined linear-logarithmic responses in photodiode (PD)-type 4-transistors active pixel sensor (APS) cells is described. we found that our APS cells can be operated with seamlessly combined linear-logarithmic responses by setting the reset transistor in weak inversion through properly selected bias voltages.

View full abstract

We propose an image-compression algorithm for micro-capsule robots to support medical diagnosis, treatment, and operations inside the body. For medical purposes, a capsule robot is required to provide doctors with high-quality images without losing important informations for medical decisions. Unfortunately, the power supply to a capsule robot inside the body and the band-width required for communications are extremely limited. Thus, we face trade-offs between power, compression-rate, and quality. To overcome these trade-offs, we analyzed the features of organ images and investigated what kind of information doctors need. The proposed algorithm achived a high-compression-rate with low computational costs without loss of information for doctors that utilized the color features and image frequencies and that were equipped with the region-of-interest mechanism. In comparison with JPEG and JPEG2000, the algorithm realizes a large reduction of computational costs and equivalent in quality concurrently.

View full abstract