Abstract: Aiming at the problem of low photorealistic models in the current image-based modeling system in the current virtual simulator, a new algorithm of image stitching and blending was studied for the images under different angles which are not in the identical plan.Firstly, use scale invariant feature transform (SIFT) to extract invariant features from images mosaic.Secondly, adopt approximate nearest neighbor searching and epipolar constraint to perform reliable matching.Then, use random sample consensus (RANSAC) to obtain the parameters of transform matrix.Lastly, stitching and blending the images.Experimental results show that the image stitching effect is wonderful and the algorithm is highly valuable in practice.