A method is proposed for solving the mapping problem from three-dimensional color space to the four-dimensional CMYK space of printer ink signals by means of a neural network. The CIE-L^*a^*b_* color system is used as the color space. The color reproduction problem is considered to be the problem of controlling an unknown static system with four inputs and three outputs. A controller finds the CMYK signals to produce the desired L^*a^*b_* values from a printer. Our solution for this control is based on a two-phase procedure. The first phase determines a neural network modeling the printer. The second phase determines the whole network system combining the printer model and the controller so as to learn an identity mapping in the L^*a^*b_* color space. The network of the controller part is then used for color conversion. Practical algorithms are presented with multilayer feedforward networks. The validity of this method is shown in an experiment using a dye sublimation printer.

View full abstract

A new method for image data representation and its application to color correction are described. In this method, an image is transformed to a representation based on Delaunay triangulation. This transformed data consists of two major parts : the Delaunay triangulation, whose mother points are feature points of distribution in color space, and a sequence of pixel values. Each pixel value is represented by coordinates based on the edges of the tetrahedron that contains the pixel. Color correction is carried out by translating these feature points and inverse transformation. In computer simulation, it was shown that the quantization error transforming our representation was comparable to the linear quantization in L*a*b* color space. In the case of 24 bit/pixel, rmse was 0.19 in proposed representation, 0.21 in the linear quantization in L*a*b* color space, 0.30 in RGB, and 0.59 in XYZ. This color correction is applied to an actual image, where no visual distortion was observed.

View full abstract

It is important to understand our color perception in memory because in our daily life we use our memory of the colors of objects when we compare them. So we studied exactly how we recognize the colors of the visual scenes that we have already experierced. A memory identification task was performed using natural scene pictures with various chromatic or luminance contrasts. Our experiment consisted of two stages, a memory stage and a recognition stage. The subjects' task was to judge, in recognition stage, if the picture was the same as that memorized. The results showed that people tend to remember a picture as having higher chromatic contrast than actually presented when they memorized it. This suggests that colors in our visual memory shift toward more saturated or categorically focal colors. This result should be taken into account when establishing the optimal method of estimating colors.

View full abstract

[in Japanese]

View full abstract

[in Japanese]

View full abstract

[in Japanese]

View full abstract

[in Japanese]

View full abstract

Because of their simple geometry, Yagi-Uda antennas have been widely used in various fields. There are a variety of techniques to analyze the near field of antennas. To examine how electromagnetic wave energy propagates around antennas, we calculated Poynting power flow. The strength of Poynting power flow is expressed by the number of arrows. In this paper, we calculated Poynting's vectors for a three-element Yagi-Uda antenna using parameters of maximum gain and maximum F/B ratio, when placed on a flat ground (wet and dry ground) and also on a perfect conductor plate. As a result, it is found that the rotating power flow can be observed in the case of specific height and above the feeding element due to the effects of reflected wave.

View full abstract

In magnetic recording, a nonlinear magnetic transition shift occurring in the recording process causes distortion of the reproduced signal. This paper proposes recording and playback equalization methods for distortion of the reproduced signal when 8-14 modulation, whose minimum reversal interval of the recording signal is a 2 codebit length, is applied. First, the amount of the nonlinear magnetic transition shift is measured using a digital VTR. Second, recording and playback equalization methods are studied by simulation based on the measured data, and are examined using the VTR. Recording equalization is a shift of the reversal points of the recording signal within the previous 3 codebit length so that the magnetic transition shift may be smallest. Playback equalization is a delay of the phase in the middle and high-frequency regions and an emphasis of the amplitude in the middle-frequency regions of the reproduced signal. By applying the recording and playback equalization methods, it is found that the height and width of eye patterns are broadened and byte error probabilities are improved by about 2 digits and 1.5 digits, respectively.

View full abstract

Computer animation is typical of artificial image sequences different from video images. Video compression standards are not well suitable for the artificial images. In order to efficiently compress animation, this paper introduces a novel concept 'three-dimensional structure runlength coding' (briefly 3D SRC), which is based on the two-dimensional structure runlength coding (briefly 2D SRC). The paper presents a basic algorithm of 3D SRC which is expanded simply from 2D SRC. In 3D SRC, the key technique is to compare each pixel of the current frame with its three-dimensional neighbors-a pixel of the previous frame and neighboring pixels of the current frame. The results of the comparison are compressed by runlength coding. Our experimental results show that in compression of animation, the data compressed by 3D SRC become one-third of the data compressed by 2D SRC on average.

View full abstract

Two experiments were carried out to investigate the required luminance and luminance contrast for transmissive liquid crystal displays. In experiment 1, 30 subjects rated 156 luminance and luminance contrast combinations displayed on a transmissive thin-film-transistor liquid crystal display, under three kinds of ambient illuminance conditions : 70, 200 and 500 lx. The results indicate that under a 500-lx illuminance condition, the combination of a display luminance approximately 100cd/m² and a contrast ratio higher than 10 is the most preferable condition. In experiment 2, 20 subjects were asked to adjust the display luminance of the transmissive thin-film-transistor liquid crystal display as they most preferred every two minutes during a visual search task, under two ambient illuminance conditions : 70 and 500 lx. The results indicate that the preferable display luminance is affected by adaptation luminance for the display screen. The range of preferable display luminance between subjects is relatively wide. These results suggest that a wider adjustable range of the display luminance is required in order to adapt a preferable display luminance for each user.

View full abstract

This paper proposes a new method to detect focal length using partial auto-correlation to improve efficiency. In partial auto-correlation, forward and rearward predictions are carried out simultaneously, linear prediction is performed for the point spread function incorporated within the defocused image, and the amount of defocusing is estimated. Because the amount of processing for the two-dimensional analysis here is enormous and high-precision calculation is required, this system uses preprocessing of the frequency dimension to extract and analyze only the horizontal component.

View full abstract

[in Japanese]

View full abstract

[in Japanese]

View full abstract