英语原文共 11 页，剩余内容已隐藏，支付完成后下载完整资料
外文翻译
Edge Drawing: A combined real-time edge and segment detector
边缘绘图：一种组合的边缘段检测器
Abstract摘要
We present a novel edge segment detection algorithm that runs real-time and produces high quality edge segments, each of which is a linear pixel chain. Unlike traditional edge detectors, which work on the thresholded gradient magnitude cluster to determine edge elements, our method first spots sparse points along rows and columns called anchors, and then joins these anchors via a smart, heuristic edge tracing procedure, hence the name Edge Drawing (ED). ED produces edge maps that always consist of clean, perfectly contiguous, well-localized, one-pixel wide edges. Edge quality metrics are inherently satisfied without a further edge linking procedure. In addition, ED is also capable of outputting the result in vector form as an array of chain-wise edge segments. Experiments on a variety of images show that ED produces high quality edge maps and runs up to 10% faster than the fastest known implementation of the Canny edge detector (OpenCVrsquo;s implementation).
我们提出了一种新颖的边缘段检测算法，可以实时运行，并产生高质量的边缘段，每个边缘段都是线性像素链。 与传统的边缘检测器不同，传统的边缘检 测器在阈值梯度幅度集群上工作以确定边缘元素，我们的方法首先沿着称为锚点的行和列分列稀疏点，然后通过智能的启发式边缘跟踪程序连接这些锚点， 因此命名为边缘绘图 （ED）。ED生成的边缘图总是由干净的，完美连续的，局部优化的单像素宽边缘组成。 边缘质量指标固有地得到满足，而无需进一 步的边缘链接过程。 此外，ED还能够以矢量形式输出结果作为逐行边缘段的阵列。 对各种图像的实验表明，ED产生高质量的边缘图，比Canny边缘检测 器（OpenCV的实现）的最快的实现速度还要高10％。
Keywords: 关键词
Edge detection 边缘检测
Edge segment detection 边缘段检测
Edge quality metrics 边缘质量指标
Real-time imaging 实时图像
- Introduction
Edge detection is a very basic and important problem, and is a popular first step in many computer vision and pattern recognition applications . A high quality edge map, which is defined to consist of perfectly contiguous, well-localized, non-jittered and one-pixel wide edge segments, significantly improves the performance of the rest of the computer vision and pattern recognition application; whereas, a low quality edge map, which has poor connectivity, edge width that varies, erroneous branching and notches, etc., significantly impairs the success of the rest of the application.
边缘检测是一个非常基本和重要的问题，是许多计算机视觉和模式识别应用受欢迎的第一步。高质量的边缘图被定义为由完美连续的，局部优化的非抖动和单像素宽边缘部分组成，显着提高了计算机视觉和模式识别应用的其余部分的性能; 而具有差的连接性，边缘宽度变化，错误分支和缺口等的低质量边缘图会显著地损害了其余应用的成功性。
A generic way to detect edges in traditional edge detection algorithms is to apply a sequence of filters to the image. The first step is usually to pass the image through a low pass filter in an effort to reduce noise and smooth out the image. Since edges are high frequency components of an image, a high pass filter (a derivative or gradient operator, e.g., Prewitt , Sobel and Feldman , Scharr, etc.) is then applied to extract out areas where edges would be located. The output of the gradient operator is called the gradient map, which is usually thresholded to suppress low frequency, non-edged areas of the image. Pixels that survive the gradient thresholding constitute what is called the edge areas, which consists of several pixel-wide, thick strips of the image. The final task is then to go over the remaining pixels in the edge areas and obtain thin, contiguous, non-jittered and well-localized edges. This is the most important and difficult step in edge detection, and is achieved by such techniques as skeletonization, nonmaximal suppression, edge thinning, or the application of morphological operators. But these techniques evaluate each pixel in the edge areas to be a potential edge element (edgel) in an isolated and individual manner. Canny edge detector, for instance, tests each pixel for being an edgel by comparing its gradient value with two of its neighbors, and if it (among a pre-set thresholded gradient magnitudes) is higher than its neighborsrsquo; gradient values along the same gradient direction, then the pixel is marked as an edgel. Such short-sighted local reasoning leads to low quality edge maps consisting individual, garbage-like edgel formations with multipixel wide, discontinuous edge segments.
传统边缘检测算法中检测边缘的一种通用方法是将一系列滤波应用于图像。第一步通常是通过低通滤镜传递图像，以减少噪点并平滑图像。由于边缘是图像的高频分量，因此应用高通滤波器（导数或梯度算子，例如Prewitt，Sobel和Feldman，Scharr等）来提取边缘将被定位的区域。梯度运算符的输出称为梯度图，通常被阈值化以抑制图像的低频，非边缘区域。在梯度阈值处生存的像素构成了所谓的边缘区域，其由图像的几个像素宽的条纹组成，然后最后的任务是通过边缘区域中剩余的像素并获得薄，连续，非抖动和局部优化的边缘，这是边缘检测中最重要和最困难的一步，通过骨架化，非最大抑制，边缘变薄或形态运算符的应用等技术实现。这些技术将边缘区域中的每个像素以孤立和单独的方式评估为潜在的边缘元素（边缘元素）。例如，Canny边缘检测器通过将其梯度值与其两个邻居进行比较来测试每个像素作为边缘，并且如果它（在预设的阈值梯度幅度之中）沿着相同的梯度方向高于其邻居的梯度值，则将像素标记为边缘，这种短浅的局部判决导致了低品质的边缘图，包括单个的，垃圾状的结构，具有多个细长的不连续的边缘段。
To improve the quality of the edge maps produced by traditional edge detectors, many post-processing techniques have been proposed. These techniques take a binary edge map as input and try to improve the edge map quality by employing rule-based methodologies with pixel-wise templates in an effort to clean up the edge map and link possible discontinuities. They also try to optimize the modal structure of edge segments with the help of morphological operations.
为提高传统边缘检测器生成的边缘图的质量，人们提出了许多后处理技术。 这些技术采用二进制边缘图作为输入，并尝试通过使用基于规则的方法与像素模板来提高边缘地图质量，以努力清理边缘图并链接可能的不连续性。 他们还尝试在形态学操作的帮助下优化边缘段的模态结构。
To better understand why edge linking and optimization studies constitute a crowded set in the literature, refer to Fig. 1, which shows the edge map for the famous Lena image produced by OpenCV Canny [4,36] algorithm, where some of the low quality edge patches have been highlighted. We see from the figure that the edge map has many low quality components consisting of ragged, discontinuous, unattended, multi-pixel wide edges, while some prominent details in the image, e.g., the left vertical line in patch lsquo;arsquo;, are undetected. Threshold values can be decreased in an attempt to detect more details, but the resulting edge map would contain more garbage-like edge instances instead of mean- ingful details. Although edge linking methods may improve the visual quality of the edge map by making use of various rules and templates, they s
全文共26369字，剩余内容已隐藏，支付完成后下载完整资料
资料编号：[143091]，资料为PDF文档或Word文档，PDF文档可免费转换为Word
以上是毕业论文外文翻译，课题毕业论文、任务书、文献综述、开题报告、程序设计、图纸设计等资料可联系客服协助查找。
