外文文献参考 使用特征值、光栅扫描算法和Hough变换相结合的方法提取线性特征

内蒙古工业大学毕业设计外文翻译

Fig1.(a) Original Imge Fig1.(b) Noisy Image

Fig1.(c) Hough transform with peaks

5. PERFORMANCE COMPARISON

In performance evaluation the correctness of output can be measured by the error rate which is the ratio of the real primitives to the numbers of spurious and mismatched primitives. Table-2 shows the error factor obtained for different HT variants and our method. From this it is evident that, our method identifies correctly the linear features. Hence the error factor is 1.0142 which is less than any other HT based methods. The graphical representation is shown in Figure-4. The computational time of our method is

5

内蒙古工业大学毕业设计外文翻译

approximately equal to the CHT but the accuracy of linear point identifications is comparatively good when compared with other Hough transform methods. The computational time required to identify the linear primitive after performing the HT and valid peak identification of HT are shown in Table-2. The computational time for Hough transform and voting process for peak identification is almost same for all the methods even different peak detection methods are used by various authors [10]. In our work the execution time is concerned with only the identification of linear segments. From Table-2 it is evident that our method based on small Eigen values approach and Bresenham's raster scan algorithm takes less computational time and it is shown in Figure 5.

Fig. 4: Error factor of different HT and Proposed method

6

内蒙古工业大学毕业设计外文翻译

Fig. 5: Computational time in seconds

6. CONCLUDING REMARKS

In this paper we discussed an efficient and accurate method for geometric primitive such as line identification in real world images. Performance evaluation is be done by comparing different HT methods with statistical parameters which uses small Eigen values and Bresenham's raster scan algorithms. The algorithm based on HT requires more memory and high computational time and will consider the elements which are not part of the required geometric primitive but having the same slope and orientation. But Guru et al [9] solved this problem by using small eigen value approach. The limitations of this approach are that the technique has to be applied for the entire image by defining proper window size and threshold values. Illingworth et al [8] solved this problem by using hierarchical approach treating the shortest line segments are the parts of long line segments. In our approach we solved all the restrictions by employing small eigen value approach for the resulting image of Hough transform. Hence by selecting the required interest Hough peak is used to determine the line segment. This reduces the memory requirement and number of computations and increases the accuracy in line identification.

7

内蒙古工业大学毕业设计外文翻译

使用特征值、光栅扫描算法和Hough变换相结合的方法

提取线性特征

在本文中，我们提出了一个新的线性几何图元识别方法，利用广义标准Hough变换（HT），基于本征值的统计参数分析和布氏光栅扫描算法。在这种方法中，我们使用稀疏矩阵找到Hough变换的图像。由于稀疏矩阵压缩零元素和含有少量的非零元素，他们提供了一个优势，在矩阵的存储空间和计算时间。基于Hough峰附近的抑制方案确定。在找到有意义的和独特的Hough峰，在Hough空间中的线性特征的坐标可以用Bresenham算法得到的光栅扫描。由于在HT参数空间量化给出真实和虚伪的原语，因为在数字图像的空间量化，在HT参数空间量化以及事实上，典型的图像的边缘是不完全的几何特征，统计分析是通过使用特征值来描述和识别的几何原语。该方法具有内存小，速度快，准确的数字化ofhough空间少的直线提取误差比以前提出的HT技术及其不变量。

1、引言

基元提取数字图像的基本任务之一是计算机视觉和图像分割。HT和它的变体，[ 1 ]是一个流行的方法提取几何图元，如直线，圆和椭圆。几十年来，许多研究者自己修改的基本HT [ 2 ]提取几何图元更有效的。基于改进的HT图像空间中随机抽样结合使用，得分积累在参数空间和收敛映射为两个空间之间的桥梁的随机Hough变换（RHT）[ 3 ] [ 6 ] [ 7 ]。在RHT可以随机抽样的像素直接从每个像素被选中的概率相等的图像。这种方法的主要缺点是，它不是几何图元组成将参与收集证据的元素，提高了计算的复杂性。在概率Hough变换（PHT）[ 4 ]，用于投票点的分数是由自组织或利用先验知识的规定几何图元。在第一步骤中的PHT点随机子集选择和标准Hough变换（SHT）[ 2 ]是随后在子集进行。为了减少计算量，进步的概率Hough变换（PPHT）[ 5 ]利用需要可靠地与不同数量的支撑点检测线票馏分的差异。这反映了在线检测过程的渐进性PPHT发现最长的线的第一和收益最短的线。一个分层的方法，基于HT Josef基特勒等人[ 8 ]开发线检测采用金字塔结构与金字塔分割成若干子图像的完整图像各层。在金字塔短线段检测逆向分层。该算法进行自下而上的分组线段到当地的社区内，较长的线。另一方面，D.S.大师等人[ 9 ]中提出的统计和几何的多连通区域支持一组边缘像素的协方差矩阵特征值的性质进行了探索，小的直线识别的目的。在该方法中的小特征值分析是用来决定一个像素的像素作为一个线性突出。

8