Image stitching has developed rapidly in recent years. Seam elimination plays a critical role in image stitching. Therefore, an improved seam elimination method of image stitching is proposed in the paper. First of all, images are registered. Then, optimal seam method based on Curvelet transformation is proposed to eliminate the seam. Objective evaluation indexes (PSNR and SSIM) are employed to evaluate the performance of the proposed method in the experimental results. A new metric of assessing the local quality of the stitched image is also proposed in the paper. Three groups of images are tested under this metric. Experimental results show that our method can eliminate the seam in an efficient way.