基于统计估计的图像边缘检测

针对传统的基于微分边缘检测算法的噪声鲁棒性较差这一问题，分析了图像边缘形成机理，依据统计估计评价准则，提出一种基于统计估计的边缘检测算法.该算法将像素点的圆形邻域沿θ方向分割成2个半圆，分别计算出2个半圆内像素样本的均值和方差，分析2个半圆内样本统计估计的有效性，进而排除噪声和纹理对边缘检测的影响.根据边缘两侧像素样本的统计估计偏度呈屋脊形分布的特点，对边缘进行细化和连接，以无噪声Canny算子检测得到的边缘作为先验知识，运用条件概率建立边缘检测评价函数.实验结果表明，基于统计估计的图像边缘检测方法具有较好的检测精度和噪声鲁棒性.

Image edge detection based on statistical estimation

In order to solve the problem that the traditional edge detection algorithms based on differential have poor noise robustness, the formation mechanism of the image edge was analyzed and an edge detection algorithm based on the statistical estimation was proposed according to the statistical estimation evaluation criteria. The algorithm divided the circular neighborhood of a pixel into two semi circles in the θ direction and calculated the mean value and variance of pixel samples within two semi circles separately. The effectiveness of the statistical estimation of the samples within two semi circles was analyzed, and the effects of both noise and texture on the edge detection were eliminated. The edge refinement and connection were completed on the basis of the character that the statistical estimation bias of the pixel samples at both sides of the edge shows a roof like distribution. The edge detected by noise free Canny operator was used as the prior knowledge, and the edge detection evaluation function was established based on the conditional probability. The experimental results indicate that the image edge detection method based on the statistical estimation has high detection precision and strong noise robustness.