Backpropagation is a one of the most typical learning methods employed in neural networks (NN). Although the method is very efficient for numerical data processing and learning, the calculation is considerably complicated. In this study we propose an alternative learning method using a genetic algorithm(GA). Our Preliminary results show that the classification rate by using the NN with GA is comparable to that with backpropagation for classification of simple categories of images. And the former method is slightly superior to the latter method for classifying difficult categories. Moreover, our results show that the proposed method has better feature in terms of convergence for obtaining the optimum solution in learning.

抄録全体を表示

An arithmetic coding is well known as a powerfull and an interesting data compression techniques from information theoretic viewpoints.
In order to apply this techniques for the multi-valued images, a simplified and a feasible processing method have to be developed. In this paper, to reduce the dependency among the neighbour pixels, and to detect the edge regions, we introduce the directional gradient by obtaining the difference value between the adjacent pixels. By combining the directional gradient on the local area, we decide whether the observation pixel is in the stationary region or in the edge region. Then a best estimate to the above pixel is obtained to compute the prediction error. After above procedures, difference (error) image which is processed by an arithmetic coding is derived. Simulation studies were carried out for girl image (SIDBA) and X-ray image to show the effectiveness of our approach.

抄録全体を表示

In this paper we present a novel algorithm, an active filter, for automated detection of ventricular contours in cardiac magnetic resonance(MR) images. The active filter, which is different from simple global thresholding methods, is a local thresholding method. Namely, the threshold values are various depending on the gray - value distribution of the local area in an image. In order to verrify the accuracy of the proposed method, some simple images are used. And then chinical MRR images are used in our preliminary experiments. The superiority of the new method is also confirmed as compared to the simple global thresholding method.

抄録全体を表示

“MTF's and Wiener Spectra of Radiographic Screen-Film Systems”, a book published in 1982 by BRH, covers methods to measure LSF. These methods have been known as the standard for MTF measurement. However, in measuring MTFs with the same sample it has been revealed to produce different values measured at different facilities. Several researches have shown that this occurs because of differences in the methods of irradiation, film processing, and density measurement methods. The authors have additionally believed that the essential cause lay in the crossover effect and low-intensity reciprocity law failure. Thus, the present study aimed to clarify how these two factors affect on the LSF measurements.The results of research have clearly shown that they are deeply involved in LSF measurement. Based on newly obtained knownledge, it is necessary to ensure experimental conditions in addition to the methods described in the book for compensation of the differences among acilities.
The results obtained were as follows:
1. With respect to the crossover effect, approx.30% of the amount of luminescence from screens affect the opposite emulsion on SR-Gfilm.
2. Low-intensity reciprocity law failure occurs with SR-G film when illuminance is 0.2 lx or below.
3. It is not possible to measure accurately LSF unless one performs the following in addition to the items described by the BRH's authors.
1) The two slit images of low or high density should be photographed under an identical exposure time with different X-ray intensities.
2) The sensitometric strip should completely be produced by the intensity scale method, and the exposure time should be the identical time with that of the slit image photograph.

抄録全体を表示

Two different methods for the modulation transfer function(MTF) measurements of screen -film systems, i. e., slit method with Fourier transform on the line spread function and square-wave response function (SWRF) method, have been widely used as a powerful tool for evaluation of the resolution property of screen-film systems. However, it is still uncertain whether MTFs obtained by the two methods are identical. Also, the nonlineality associated with double emulsion film of highly asymmetric screen-film systems must be considered in their analysis. One of the recent work concerning MTF measurements of screen-film systems is the standardization of MTF measurement. The standardization of the MTF measurements have to be made based on a consensus between standardization organizations, such as American National Standards Institute (ANSI) and German Standards Institute (DIN). Investigation for standardization of MTF measurements of screen-film systems in ANSI has been carried out at the University of Chicago in 1991-1993. In our study, we showed several important results regarding the SWRF method, which is considered relatively unknown compared to the slit method. In this article, we will mainly describe the technical factors affecting the accuracy and the reproducibility of MTF measurements with the square-wave response function (SWRF) method. Furthermore, we will discuss about the comparison of MTFs obtained by the two methods based on the interlaboratory comparison of MTF. Finally, we will describe the problem of the nonlineality associated with double emulsion film of highly asymmetric screen-film systems.

抄録全体を表示