The butt weld is common in the friction stir welding. The assembly quality of the weld， which could have a great influence on the joint performance， is generally described by gaps， misalignment and non-centering of probe （NCOP for abbreviation）. Linear structured light is a common method for extracting the characteristics of welds. Nowadays， the contour information of the weld is mostly represented by discrete points based on linear structured light. How to efficiently extract the contour information of the weld from the contour points becomes a new challenge. This paper proposes a method for evaluating weld assembly quality based on machine vision， which converts discrete contour points into bitmaps， improves the recognition reliability of line features by using anti-aliasing algorithms， Then， the assembly quality information is acquired for weld assembly quality evaluation. Compared with the traditional discrete point fitting method， this algorithm has obvious efficiency advantages.