Study On Seamless Image Stitching Based On Curvelet Transformation

Image stitching is a process that combines images with overlapped areas altogether to form an image with large view and high resolution.In recent years,image stitching has been widely developed and applied.In this work,by reviewing the former works,a stitching seam elimination algorithm which combines with Curvelet transformation and optimal seam method is proposed.In our algorithm,the cost matrix is formed by calculating and weighting pixel and gradient values under two registered images.Then,shortest path algorithm is applied to search the optimal seam.Next,the overlapped areas are transformed into Curvelet domain by Curvelet transformation.Different rules are applied to high frequency and low frequency parts of the image.Finally,inverse Curvelet transformation is taken to transform the fused image into spatial domain and put into the stitched image.In the experiment part,three sets of images are taken.After subjective assessment,peak signal to noise ratio(PSNR)and structure similarity index(SSIM)are taken for objective assessment.To carefully assess the quality of stitched images,experiments are designed to select 5,25 and 50 pixels both sides near the stitching seam.The quality is assessed using the gradient variance.Experiments show that selecting 5 or 50 pixels cannot fully meet the assessment requirements.However,the assessment requirements can be met by selecting 25 pixels both sides near the stitching seam.To further continue our experiments,we resize those images to scales of 50%,80%,120%,150% to validate our criteria.This assessment criteria fills the blank of the assessing of stitched image near the seam.The results show that our criterion is suitable for assessing stitched images and that our algorithm has advantages in seam elimination in image stitching.