In order to solve the image content and single feature points is not obvious and the number of little and lead to its difficult to splice the exact problem, put forward the image stitching algorithm based on probability and matching feature points. Should first, using the image characteristics of the overlapping area, developed for rotation, scale and illumination invariant feature geometry structure, reliable calculating distance matrix and the rotation matrix, realize the feature point detection; Using random sampling consistency algorithm, feature point matching. Then, using the Bernoulli distribution characteristics, and calculation of the bayesian model points and outliers in probability, eliminate error matching points, so as to improve the image matching accuracy, accurate finish image stitching. Last barcode docking is accurate as benchmark image splicing quality judgment, experimental test results show that compared with the current image splicing technology, the joining together of this algorithm has higher accuracy and robustness.