为了提高光学零件表面划痕定位测量和宽度测量的精度，采用基于离散正交多项式曲面拟合的亚像素边缘检测与宽度测量的算法，通过对边缘点及其邻域进行曲面拟合代替了只对边缘点梯度方向进行曲线拟合，实现了亚像素边缘检测；采取感兴趣区域加速策略与基于离散正交多项式曲面方程参量快速求解方法串联进行的加速方案，大大减少处理时间；在宽度计算方面，根据划痕长度自适应分段，分别计算每段分段点之间欧氏距离并作为其宽度，比较后取最大宽度作为划痕宽度。结果表明，该算法测量精度较高且具有较强的鲁棒性，同一划痕在不同视窗下测得宽度误差均值不超过5.2%，且标准差不超过0.3；在求解曲面模型参量的时间方面，该方法计算时间约为最小二方法的7.35%，处理效率显著提高。该方法能够满足工程应用中快速、高精度的测量要求。

In order to improve location and width measurement precision of optical surface scratches, a sub-pixel edge detection algorithm based on discrete orthogonal polynomial surface fitting was proposed. Sub-pixel edge detection was realized by fitting the edge points and their neighborhood into surfaces instead of by curve fitting only in the gradient direction of the edge points. By using the acceleration strategy of the region of interest and the method of fast solving surface equation parameters based on discrete orthogonal polynomial, the processing time was greatly reduced. In terms of width calculation, the Euclidean distance of each section was calculated after the scratch was divided into a series of small segments adaptively and then the maximum width was the width of scratch. The experimental results show that the proposed method has high accuracy and strong robustness. Mean error of width is less than 5.2% and standard deviation of width is less than 0.3 of the same scratches measured under different visual windows by this algorithm. In terms of processing time to solve the surface model parameter, the processing time of acceleration strategy is 7.35% of the least square method and is greatly reduced. This method can meet the requirements of fast and high precision measurement in engineering application.