We developed a wide-band, high-efficiency, cost-effective multi-channel power amplifier (MCPA) using digital pre-distortion for digital terrestrial TV broadcasting. In this equipment, a wide-band of 50 MHz and over, is produced by using a high-speed A/D D/A converter that has a sampling frequency of 200 MHz and over. We also developed a new method that detects PA distortion by adding an RF output signal and an inverted RF input signal to reduce dynamic range and used a new algorithm that compensates for a high speed.Using these new methods, we obtained IMs of-50dB and under with multiple channel input and high efficiencies of 10.2-11.2%.

View full abstract

We examined adaptability of a mechanism for stabilizing vision in a conflicting environment, where visual information and head motion were not coherent. We defined “Gain” as the angle ratio of amplitudes of visual motion to those of head motion. First, the subjects adapted to a virtual environment in which they were surrounded by either a sphere or a cube stimulus. The Gain was set at either 0.5 or 1.0 with the head motion following a moving sound for 2 minutes. After this adaptation phase, the subjects adjusted the Gain to obtain the best visual stability. The results showed that the visual stability mechanism starts to recalibrate to such incoherent environments quickly, regardless of the stimulus pattern. Moreover, the results suggested that the visual stability mechanism des not independently process along the yaw and the pitch axes.

View full abstract

Extensive studies have been done on manipulating individual facial attributes, such as expression and age, to synthesize facial images. Most studies have focused on manipulation of the shapes of facial parts, such as eyes, mouths, or outlines, but spots and wrinkles have been treated as simple global features, such as skin color and texture. This research, on the other hand, analyses the shapes of individual spots and wrinkles for their extraction and manipulation. Extraction and classification of the wrinkles and spots are based on the characteristics of pixel distribution within each spot pr wrinkle, not on its contour shape. We conducted human evaluation experiments, confirming that manipulating spots, wrinkles, and facial parts altogether more effectively manipulates age than manipulating only facial parts.

View full abstract

Adding color to a grayscale image is important in various fields. However, much trial and error is needed to obtain the appropriate colorized result since no standard method exists for this procedure. We propose a method to colorize grayscale images using a neural network. The relation between brightness and color in a reference color image was studied using the neural network. The grayscale image was then colorized through the network. For images consisting complicated scene components, multiple neural networks among the swatches proposed by Reinhard et al. were utilized. We show the superiority of our approach over the conventional methods by demonstrating several examples.

View full abstract

In the broadcasting industry, how to create contents is currently being discussed. Most broadcasting stations in Japan are worried about a shortage of contents broughton by the introduction of disital broadcasting. We have developed an approach that helps creators form new contents ideas. An approach to support the need for contents in the program production. For our approach, we used a method called “Knowledge liquidization and crystallization”, and a system called the “Knowledge Nebula Crystallizer (KNC)”. To investigate the cyclical creation process. To begin with, we collected of various scripts and examined them. We analyzed how to work for supporting the process of knowledge creation in the program production. As an example knowledge creation, we have applied our approach to actual radio program design processes.

View full abstract

We propose applying of the ridgelet transform to semantic objects segmented using watershed segmentation. The ridgelet transform effectively represents objects with singularities along lines; therefore, it is a powerful tool for coding. Moreover, it can easily rotate blocks in the ridgelet domain. Nevertheless, the targets to be roteted are not rectangular blocks but real objects. For that reason, we divide pictures into semantic objects using watershed segmentation, then convert each object into the ridgelet domain so they can be rotated easily. We experimented in coding and rotating objects.

View full abstract

The parallax barrier display system has excellent characteristics. For example, no special glasses are required. However, a conventional multi-view parallax barrier display system has disadvantages such that the resolution of each of the stereo images is reduced only horizontally. This paper describes a 4-view display with a parallax polarizer slit and cylindrical lenses. The advantage of proposed method lies in its ability to distribute resolution problem both to the horizontal and the vertical planes, then this three-dimensional (3-D) display avoids the horizontal resolution problem of conventional system.

View full abstract