A novel method for medical image fusion is proposed. It can be used to define lesion focus during treatment planning. Edge detection is studied by using an improved Canny edge detector, and threshold for non interesting region is made by calculating local image histogram. Space mapping is built by image registration. Features of lesion focus in an image can then be overlaid with other corresponding images and a fused image with features of edge of lesion focus and anatomic structure can be obtained. Results of experiments showed that the proposed method is simple and intuitive, and can be used in clinical application very well.

To realize virtual simulation and virtual radiograph, a fast ray tracing algorithm for digitally reconstructed radiographs (DRRs) is presented. Calculation of the absorption of X-ray by the tissue employs the CT data. Bresenham′s two dimension line generating algorithm is generalized to three dimension for X-ray tracing. Without calculations for equation solving, only +/- operation for calculation of interaction points is included. The algorithm is proved to be fast and shortens the time of image construction.

为了得到高质量的矢/冠状面(Sagittal/Coronal)重建图像,提出一种基于图像形态尺度变化的插值重建算法.对整型B样条曲面引入控制点权重,并将权重作为图像值的函数,利用B样条曲面的局部支撑性,由图像值对插值曲面的局部形状进行控制.该算法充分考虑了断层间解剖组织分布形态尺度变化的因素,因此计算精度高,插值图像具有较高的保形性,断层间插值层的过度连续性为C(n)(n>1).

Based on results of Freeman′s physiological and simulation experiments, we summarized Freeman and Tsuda′s explanation of chaos dynamics in bio neural network. From point of views of information flow, we discuss the necessity of existence of chaos in bio neural system, we discuss the potential information processing capability of models in ANN, where chaos mechanism has been introduced. We point out that, when there is input from outside world, the reaction of a cognition system is a change of characteristics of it′s dynamics behavior, not only static output value. It is also suggested that neural networks with characteristics of chaotic dynamics have ability of pattern classification and pattern interpretation simultaneously. This combination comes from characteristics of chaos dynamics of the system, describing the role of chaos in pattern recognition. We point out some advantages, comparing our model with those traditional neural network models like ART. Part of the conclusions were proved during simulation experiments.

We proposed a model that utilizes oscillator neural networks to encode perception information. Based on this model, we find a method to perform multi-resolution pattern analysis. Because the oscillators are interconnected, the local features of pattern are bound in a natural way. Conclusions are proven by computer simulation.

A novel approach of progressive transmission based on wavelet descriptor for surface rendering is proposed. The method presented in this study can be used as a transmission means to apply network for medical data “visualization”. In the first part of the paper, the digital contour wavelet descriptor is defined focusing on features of the series of plane contours used for surface construction where the contours are periodical ones. The basic idea of progressive transmission based on wavelet descriptor is as follows. There are three steps for the progressive transmission procedure. The first one is to represent contours using the wavelet descriptor at the sending site. Then the coefficient series of the wavelet descriptor are transmitted progressively. At the receiving site contours were reconstructed using the descriptor coefficient series, and the surface was reconstructed and rendered. Reconstruction accuracy and rendering effect were refined progressively with data of details, and were improved to image quality at the sending site finally. Transmission can be of full fidelity and bandwidth demand could be moderated. Experiments provided in this paper also show the feasibility of this method.

为了描述数字空间目标轮廓,便于计算机图形表达,提出一种适于描述序列点轮廓的小波描述子.将数字空间轮廓视为周期序列,并基于离散小波变换给出序列点轮廓小波描述子的具体表达.结合具体医学图像轮廓进行相应的应用研究,并与Fourier轮廓描述子进行了对比,研究结果表明,在具有相同比特率的情况下,序列点轮廓小波描述子具有更好的保形性,同时实现了对轮廓数据的压缩.

The uniqueness of the weight vector for the self-consistent sensory-motor learning algorithm and the algorithm's convergence is investigated. The effect of the choice of the local receptive field's parameters to sensing noise, intrinsic unit noise, and target function is discussed. An adaptive strategy to choose the local receptive field's parameter is suggested.<>

The non linear and random characteristics of chaos provide an effective approach to enhance information security. A perfect block cipher scheme with cipher block chaining model is described for a symmetric encryption algorithm based on the chaotic attractors in discrete Hopfield neural networks. The method is simpler and more concise than DES and some kinds of attacks can not easily unfold the scheme.

Main problems in the research of pattern recognition, forexample, feature selection, structure recognition, limited samples and ill-posed problems in statistical recognition are discussed from point of view of intelligent information processing. It is proposed that, instead of taking neural networks (NN) as a tool for implementation of existing methods in pattern recognition, we shoud pay our attention to the solution of main problems in pattern recognition.

A novel planer contour reconstruction algorithm based on fractal interpolation is presented. With the intrinsic property of fractal geometry, the algorithm can be used for describing irregularity of planer contours and representing the shape between two planer contours. The new interpolation method is achieved and applied to medical images experiment. The study results show that the proposed approach can preserve the shape of the planer contours accurately, and the amount of computation and storage is very little.

High quality coronal and sagittal images are needed to help doctors accarately locate the structures and locations around organs and focal lesions in patients for clinic diagnosis and treatment. A set of CT/MRI images were interpolated using the fractal geometry interpolation algorithm based on the iterated function system (IFS). The vertical scaling factors of the IFS were estimated according to the local box dimension of the sample data set. The experiments showed that the texture of the interpolated images can be retained with high resolution. There are many advantages of the algorithm, i.e. reducing memory and time; getting precise images. The algorithm can also be applied for interpolating general grey level images.

对称性在日常生活中广泛存在，对一般物体镜像和旋转对称性的快速检测一直是计算机视觉的难题，至今未有通用性很好且效率较高的算法．本文提出一种新的方法，把对称性检测问题转化为协方差矩阵的特征值分解问题，文中的３个定理及其证明奠定了理论基础．此外还把这一结论推广到旋转对称性的检测问题．仿真表明：这种方法理论基础扎实，操作简便，非常适合于强镜像对称复杂图像的镜像对称轴方向检测．

Automatic face recognition (AFR) is one of the most attractive and challenging tasks in fields of computer vision and pattern recognition, the first critical step of AFR is face detection. A fast face detection algorithm based on regional feature is proposed. In contrast with the award winning mosaic method, three main improvements were introduced here: 1). organ based blocking scheme and more intuitive mosaic rule design, 2). sub block type adaptive technique according to different face shape, 3). fast organ based rough detection and hierarchical local searching method. Also directional gradient statistics were used as sub block feature instead of absolute gray value ones. Experimental results show that the algorithm can find the face rapidly with relative high accuracy, with little limit to the complexity of background, lighting condition, face size, person number, resolution of the image etc.

本文从函数逼近的观点，讨论了样本量有限的重要性，简单回顾了学习理论的发展及其与神经网络的关系，指出学习理论对多层前向网络研究的重要性．

针对Ｋｏｈｏｎｅｎ的自组织特征映射（ＳＯＦＭ）神经网络的不足，本文把进化计算的思想用于神经网络的结构寻优之中，提出了一种结构自适应的自组织神经网络（ＳＡＳＯＮＮ）模型．ＳＡＳＯＮＮ基于把每个神经元看成是一个进化群体中的一个个体的观点，构造了神经元生长（ｇｒｏｗｉｎｇ）和删除（ｐｒｕｎｉｎｇ）的准则和方法，使得ＳＯＦＭ中的神经元欠利用，神经网络映射欠准确，以及映射的边缘效应等问题得到很大程度的改善

本世纪末数字地球概念的提出具有重大的意义，文章首先讨论了它的特点，接着对网络的人工生命现象进行了分析，提出数字地球神经系统模型的新概念，指出网络工操作性是数字地球神经系统的关键因素，并研究了基于地理空间信息与知识的智能代理技术。

从计算机网络应用中的信息安全出发，结合国内外计算机安全的现状，针对数字地球的特点，探讨了数字地球信息安全策略、构建原则和体系结构，并重点分析密钥和信息加／解密等现代密码学技术在数字地球系统中的应用。