图像边缘检测是数字图像处理中的重要内容之一，是图像理解和分析的基础。 本文首先介绍了图像边缘检测技术的研究背景，研究意义以及国内外当前的研究状况。阐述了数字图像处理的过程及研究方向，以及一些重要的基本概念，引出了图像边缘检测的重要性。 其次，对图像边缘检测的各种经典算子做了总结和分析，如Roberts算子、Sobel算子、Prewitt算子和Laplacian算子等。并通过理论分析和VC环境下实现效果做了比较，了解他们各自的优缺点及适用性；本文也对当前国内外针对各个经典算子相关的改进方法作了简要介绍。 随后，叙述了传统Canny算法及实现效果。针对传统Canny边缘检测算法自动化程度不高，计算梯度方向局限性的缺点，提出一种自适应调整高斯平滑滤波和多方向计算梯度幅值相结合的算法。改进的Canny算法通过空间尺度因子自适应调整的高斯滤波器对原始图像进行平滑滤波，保留完整的边缘信息的同时去除噪声的影响；另外在计算梯度幅值的时候采用了多方向邻域的梯度幅值计算方法，考虑了像素对角线方向的梯度计算，进一步抑制了噪声的影响，提高了各方向上边缘的精确定位。 最后，理论和实验结果表明，改进的Canny算法简单易实现，满足了传统算法的最优化准则，并且提供了图像边缘的自动化检测能力和全方位的边缘定位性能，具有较好的检测精度和准确性。

Image edge detection has become one of the most important parts of image processing. It is also the basis of image analysis. In the thesis, first of all, the background, the significance and also the development status of the image edge detection technique are introduced. In addition, the process and direction of image processing research is also described. As the basic concept of image processing is discussed, we know the most importance of the image edge detection. Secondly, some classical edge detection algorithms such as Roberts, Sobel, Prewitt and Laplacian are discussed. Theory analysis and experiments are carried out to compare their advantages and disadvantages. Also some improved methods for the classical operators are simply introduced after each one in this chapter. Then, the traditional Canny edge detection algorithm is lack of the automation process capabilities and limited orientations from calculating gradient in the edge detection. On the basis of the traditional Canny edge detection algorithm, a new method of scale-adaptive Gaussian filter with multi-directional gradient together is proposed as the improved Canny algorithm. The new method smooths the original image with scale-adaptive Gaussian filter, which retains the edge information and removes the impact of noise. In addition, according to gradient direction on each point, the new method can reduce noise and improve the egde location precision. Finally, with theoretical and experimental results, it is demonstrated that the improved Canny algorithm is simple and easy to realize, fulfilling the optimization criteria of the original algorithm. Moreover, compared with the traditional algorithm, the new method provides the automation process capabilities and shows the better accuracy and location precision in the edge detection.