A real time image processing system composed of a neuromorphic silicon retina and FPGA circuits was designed. Using the system, target tracking was carried out under indoor illumination. In the tracking operation, the silicon retina distinguished a moving object from background objects by subtracting consecutive image frames after smoothing. Then, the image of the moving object was binarized and the medians of the object's horizontal and vertical histograms were found. The system was found to be able to execute robust image processing under indoor illumination using compact hardware and low power consumption. Therefore, it is considered to be suitable for various engineering applications such as robot vision.
One of the problems of conventional 3D displays is the danger of using the ocular function. Observers feel fatigue because accommodation responses do not correspond to convergence. We have already proposed a new 3D display technique to solve this problem by displaying high-density directional images. This technique can be implemented by using modified 2D aligned telecentric optical systems. This technique enables several people to observe 3D images simultaneously without wearing 3D glasses and also offers color and moving 3D images. In this study we measured accommodation responses to a prototype 3D display which can display 64 directional images simultaneously. Measured accommodation responses to the prototype 3D display were very similar to those obtained for real objects.
We have proposed a natural 3D display which is based on high-density generation of directional images.In order to display photorealistic 3D images of real 3D objects, the most promising 3D camera technique is the generation of high-density directional images from multiple images captured using multiple horizontally aligned cameras by an interpolation algorithm.In this study we examined four simple interpolation algorithms. An algorithm that utilizes one representative object distance as a priori was found to be the most effective one. This technique enables a fast interpolation. There is, however, a tradeoff between the number of cameras and the allow-able depth of 3D objects. This tradeoff was examined theoretically and experimentally.
In OFDM transmission, a loss of sub-carrier orthogonality due to Doppler-spread channels leads to inter-carrier interference (ICI). This problem is especially severe in mobile reception environments. ICI causes significant degradation in bit error rate characteristics and its influence increases as the carrier frequency or velocity of the receiver increases. We propose a method to reduce ICI caused by the Doppler-spread channel. The transmission channel is modelled by a combination of multiple Doppler-shifted propagation paths and their parameters, such as attenuation, relative delay, and Doppler-shift, and are then estimated using scattered pilot symbols. The ICI is cancelled by multiplying the inverse matrix of the estimated channel matrix by the received OFDM symbol vector. However, the ICI canceller emphasizes a noise component, so it does not improve the bit error rate characteristics sufficiently. In this paper, we propose an ICI canceller based on MMSE that does not emphasize a noise component. Using computer simulation, we confirmed that ICI cancellers based on MMSE can reduce the bit error rate more than ICI cancellers based on zero-forcing under the Doppler-spread channel.
Orthogonal Frequency Division Multiplexing (OFDM) signals consist of many independent subcarriers and have a high peak power. The Partial Transmit Sequence (PTS) scheme has been proposed as an effective scheme for reducing the peak power. The scheme employs phase rotations of subcarriers, and a set of rotated angles corresponds to side information, which is transmitted through the same channel as data or another channel. In this paper a coding technique is considered to improve the performance of both data and side information in an OFDM system with PTS. Since the side information is used at the detection of OFDM, the side information should be recovered before the detection. This is, in general, against the causality, because the side information is not obtained before the detection. We proposed a new coding method based on a rotationally invariant trellis coded modulation for the data and side information. If we insert side information into the position that does not suffer from phase rotation, it becomes possible to correct the error due to rotations after decoding. Thus we can obtain coding gain and peak power reduction in the OFDM system using a PTS. As a result we show that the bit error rate of the encoder is at most twice worse than the normal.
We propose a new interline CCD (IL-CCD) image sensor which combines buried photodiodes and CCD registers driven through a barrier (DTB-CCD). The performance of the image sensor was simulated by three-dimensional numerical analyses, emphasizing on dark current and charge capacity. It was clarified that highly biased electrodes of the DTB-CCD absorbed most of the generation-recombination (g-r) currents at Si-SiO2 interfaces beneath their electrodes and also the currents between electrodes with low biases and electrodes with high biases. The g-r currents were reduced by several orders at the interface under electodes with low biases, because holes were introduced under the interface. Most of the reduced g-r currents are also absorbed into their respective electrodes. However, a small part of the g-r currents generated at the above three interfaces flew into a channel, going over the potential barrier between the bottom of the SiO2 layer and channel, to become a dark current. When the barrier height was increased, dark currents were significantly reduced. Therefore, the IL-CCDimage sensor enables a device with very few dark currents. When a 8.3μm (H) ×12μm (V) pixel includes two photodiodes and four transfer electrodes was used, more than 4.5-5.0×104 electrons were transferred from photodiodes to CCD registers by 0V to 9.5V and -5.8 to 0V transfer pulses. Additionally, 4.0×104 electrons were transferred in the CCD by -5.8V to 0 V amplitude using conventional 4-phase driving pulses.
Restoration and demosaicing methods for pixel mixture images are described. Pixel mixture is a function for saving readout time for digital still cameras, which mix together two pixels on a CCD. Pixel mixture reduces the number of pixels to be read out from the CCD, and it enables a video rate readout from DSCs of more than a million pixels. An iterative method is presented that can recover the images. Our approach achieves a smaller mean square error than conventional interpolation methods. A color correlation method is also presented for reducing pseudo color.
An anisotropic extension is presented to a bilateral filter which is an edge-preserving nonlinear filter. Spatially anisotropic weights are adjusted on the basis of the gradient tensor of the luminance at each pixel of an image. This anisotropic bilateral filter is applied to enhance stripe patterns such as fingerprints and nonphotorealistic rendering for generating flow patterns in pictures. Extension of the low-pass characteristics to a band-pass ones strengthens the stripe enhancement effect of the filter.