图像

A method based on linear singularity analysis of image is presented. The improved method, called Directional Beamlet Transform is better than traditional Beamlet transform about directional information of image and has the ability of representing linear-singularity of image. According to the analysis of linear singularity, edge detection is transferred to detecting singularity point in the coefficient matrix of DBT.

针对基于图像像素点分析的边缘提取方法存在无法同时满足高抑噪性、连续性、定位性等问题，本文提出了方向Beamle 变换方法，在定义图像线奇异性的理论基础上，利用DBT 对图像进行线奇异性分析，依据Beamlet 变换具有的线段提取能力，将图像边缘检测问题转化为方向Beamlet 变换系数矩阵中奇异点的检测问题，以降低噪声点对边缘检测结果的影响。

Methods There were 100 cases in normal group and 36 cases in group of chest injuries. All cases were performed in volume scan according to conventional chest scan by SIEMENS Sensation 4 MSCT , then performed in thin slice low and high contrast image reconstructions. After that, all the source images were input into CT 3D work stations,costicartilage were imaged by post process software such as multiplanar reconstructions, maximum intensity projection, surface shade display and volume rendering technique. All the pictures were observed and analyzed by two radiologists.

利用西门子Sensation 4 多层螺旋CT机按照胸部常规扫描条件对胸部外伤组36例和对照组100例患者进行容积扫描，然后进行薄层低对比及高对比图像重建，并将重建图像导入CT三维（3D）工作站，利用多平面重建成像、最大密度投影、表面遮盖法成像及容积成像技术对图像进行后处理，由2名CT诊断医生一起对各种后处理图像进行观察和分析。

If the setting is too low, scanned images granula edge rough, radiofrequent denticular, quality is poor; if it is set too high, you can make to the original copy in the unnecessary details, such as by blotchs, fold, and additional background around the image exposed, may also make the scanned images of storage space is too large, affecting the scan speed.

如果撤销太矮，扫描的图像颗粒粗劣，图像边际呈锯齿状，质量很差；如果撤销太矮，则会使草稿洋不必定的梗概，如画背上的斑不背、褶皱以及图像边际的其他配景突显不入来，还会使扫描图像的保存不时间过不小，感化扫描速率。

The texture feature of a picture is one of the most significant features of it, Fractal Dimensions of a picture is a effective method to descript the texture feature of a picture.

图像的纹理特征是图像的重要特征之一，图像的分形维数是描述图像纹理特征的一个有效方法。

The paper solved the problems of less valid data in thermal infrared imager,impurity being on surface of specimen and uneasy dipartite corrosion area in image which happened in detection of structure parts' hidden corrosion by thermal wave imaging.Firstly,the contrast of one unheated image was enhanced by self-adaptive mathematical morphology sharpening,and then location of the impurity was confirmed by change of the image grads.

针对飞机结构隐藏腐蚀热波成像检测中出现的试件表面存在杂物、热像仪中有用数据较少、图像序列中腐蚀区域不明显等情况，首先通过对加热前的一幅图像进行自适应形态学锐化处理，提高图像的对比度，并根据图像梯度的变换情况确定出杂物的位置。

Blood speckle noise in the IVUS image would affect the validity of edge detection badly. A temporal/spatial filtering method was used to preprocess the IVUS image to restrain the noise. Then a method to automatically detect the intima and ectoblastic edge of IVUS image was presented. This method is based on the active contour model and the contrast properties of IVUS image, which makes use of the Hopfield NN and simulated annealing algorithm.

为解决血管内超声图像中严重的血液斑点噪声影响边缘提取方法的有效性问题，采用一种时/空滤波方法对IVUS图像进行降噪预处理，基于活动轮廓模型和IVUS图像的边缘对比度特征量，利用Hopfield神经网络并结合模拟退火算法自动提取IVUS图像的冠脉血管壁内、外膜边缘。

We mainly discuss three types of inpainting models: the Bayesian model, geometric models and PDE models with emphasis on investigating TV inpainting, elastica inpainting and Mumford- Shah inpainting of PDE models. After studying the level set method and its applications to image processing, we concentrate on discussing the inpainting model...

在介绍水平集法及其在图像处理方面的应用之后，进一步深入地讨论了一类图像修复的变分模型—Mumford-Shah-Euler模型，在此基础上利用水平集法，提出MSE模型的数值逼近的算法格式，并对一些具体的图像修复问题开展数值模拟和修复，深入分析了MSE模型及水平集法在图像修复中的优缺点，这些结果在理论和实际应用中都很有意义。

Third, it has a discussion on the development of modern and traditional artificial neural network technology and their application in image processing. It summaries that the model of the Pulse-coupled neural network based on Elkhorn's model of the cat visual cortex was constructed according to the mammalian vision systems. The nonlinear modulation property, which is a generic and pervasive biological pulse-coupling mechanism, is used in the construction of PCNN. So, PCNN finds many applications in image processing, including segmentation, edge extraction et al. As all known, the performance of the image segmentation depends not only directly on the adjustment of PCNN parameters and the statistical properties of image but also on the cyclic iteration times N of PCNN. If the parameters have been properly set, it turns out to be essential to select a suitable criterion to determine N.

回顾了神经网络技术的现状，指出传统人工神经网络只利用了生物神经元有限的属性进行模型构建，而90年代产生的脉冲耦合神经网络充分利用了哺乳动物视神经网络特有属性，其神经元模型除了考虑生物神经元特有的线性相加、兴奋与抑制属性外，还考虑到生物神经元特有的非线性相乘调制耦合特性；考虑了哺乳动物视神经系统视野受到适当刺激时，相邻连接神经元同步激发产生35Hz～70Hz振荡脉冲串特性等，所以PCNN网络模型更靠近真实生物视觉特性，特别是哺乳动物视神经网络，特别适合计算机图像处理应用，尤其在生物细胞图像分割、图像平滑、降噪及图像中目标的分类、计数等方面的研究。

A new algorithm of low bit rate image compression was proposed. Firstly, the edge informations of the input image were detected with SUSAN operator, then the detected edge informations were en hanced with edge enhancement method. Finally, the enhanced edge informations were added to the decoded input image to form the final compressed image.

提出一种新的低比特率图像压缩算法，该算法首先利用SUSAN算子检测出输入图像的边缘信息，然后利用边缘增强方法对边缘信息进行增强，最后将增强后的边缘信息和解码后的输入图像相加形成最终的压缩图像。

The potential text margin in the perspective document image is a convex quadrilateral and in the oringinal document is a rectangle.The relation between the convex quadrilateral and the rectangle was used to determine a homographic matrix.

由于目前的字符识别技术大多只能处理近似无变形的文本图像，这就需要将透视变形的文本图像校正为近似无变形的原文本图像的正视图，得到现有字符识别技术能够处理的文本图像。

The split between the two groups can hardly be papered over.

这两个团体间的分歧难以掩饰。

This approach not only encourages a greater number of responses, but minimizes the likelihood of stale groupthink.

这种做法不仅鼓励了更多的反应，而且减少跟风的可能性。