In order to solve the problems caused by uneven illumination of the curved surface, such as large brightness difference and unclear feature of pipeline weldimage, a fast and stable weld image mosaic method was proposed.Based on the image mosaic algorithm with traditional speeded up robust features(SURF), in the feature description link, the local difference binary(LDB) descriptor was used to describe the feature point area, by which the feature dimension is reduced compared with the traditional descriptor.Combined with LDB descriptor, the constraint information of feature points and adjacent feature points was constructed. The local structure information was used to match multifeature point simultaneously to replace the single feature point matching method to improve the matching stability.The experimental results show that the method effectivelyimproves the speed and success rate of pipeline weld image mosaic. The efficiency of the mosaic is about 1.8 times of SURF algorithm, and the success rate of the mosaic is about 1.3 times of SURF algorithm.The method also has certain application value in other image mosaic fields