中文摘要：

在大口径精密光学表面疵病检测中,针对原有图像拼接算法在子孔径图像数目较多时易产生拼接错位的情况,提出了基于特征分类的多层次图像拼接算法.该算法分为特征分类和多层次拼接2部分：前一部分通过重叠区域内特征的提取和分类,将所有子孔径图像的重叠区域分为4类;后一部分针对4类重叠区域的不同特点,进行4轮拼接,最终得到全景图像.结果表明,该算法能有效避免由导轨累积定位误差和缺陷贯穿数个子孔图像的贯穿型直线特征导致的拼接错位,且较原有算法效率更高.因此,该算法对于实现大口径光学元件表面疵病的高分辨率检测具有重要意义.

英文摘要：

In the field of defects evaluation on the large fine optics surface,a multi-cycle image mosaic algorithm based on feature classification was proposed on account of the mosaic mismatch which happened easily when the previous mosaic algorithm was used in the situation of large quantity of mosaic sub-aperture images.The algorithm was constructed by two parts： feature classification and multi-cycle image mosaic.The former part divided overlapped areas of the sub-aperture images into four types based on extraction and classification of features in overlapped areas.The latter part conducted four cycles of image mosaic to get the final panoramic image according to different situations of the four types of overlapped areas.Result shows that the algorithm effectively avoids mosaic mismatch caused by accumulation locating error of guide rail and run-through line features which a long scratch defect run-through several sub-aperture images with higher efficiency compared to the previous mosaic algorithm.Therefore the algorithm has an important significance in the field of high-resolution evaluation of surface defects on macroscopic optics.