Image Mosaic Based On Improved Image Registration

Image Mosaic is one of the digital image processing technology. It’s mainly used to stitch a series of perspective images with the same overlapped scene to a new image of large wide viewing angle which involves all the information in the previous images, through the way of image preprocessing, image registration and image fusion. Image Mosaic is of great importance in real life. Its core technology is image registration, whose key steps are feature extraction and feature matching.This paper mainly studies the image mosaic based on improved image registration. The mian works in four aspects are as follows:(1) As is known to all, the SIFT algorithm needs to search26spots each time and compare all of them in order to judge out whether there is an extreme point. In addition, SIFT has edge response points, large dimensions of feature descriptor an so on. Therefore, it comes to an improved approach based on SIFT feature extraction. Firstly, it judges if there is an extreme point using one-stratum comparisons. Then, the approach uses Canny algorithm to eliminate DOG edge response points. Finally, it reduces the dimensions of feature descriptor from128to90. The results of this study show that the improved searching method of the extreme point increases the efficiency. That the Canny algorithm eliminate the edge response points extracts effective features through. What’s more, it successfully reduces the calculation amount by the reduction of the dimensions. The final improved SIFT approach extracts effective features and raises the processing speed at the same time.(2) Because of the low accuracy in feature matching with fixed ratio thresholds, the approach presented here is an improved adaptive ratio thresholds feature matching method which uses repetitive rate to calculate the accuracy of feature matching. The result indicates that the matching accuracy of the improved approach reaches to ninety percent or more.(3)This paper introduces the theory of RANSAC algorithm and uses this method calculate the initial transformation matrix. To achieve the seamless mosaic of translation, rotation or scaling images, it adopts a projective transformation matrix with eight parameters. Then it reduces the projection error using L-M algorithm to reevaluate the initial transformation matrix. This paper also introduces the fading-in-and-out weighted average image fusion algorithm that attains the fusion of mosaic image.(4) This study accomplished the image mosaic based on improved image registration and the image mosaic based on original image registration. The image mosaic based on original image registration:Using the original SIFT algorithm extracts the features, and achieves the feature matching with the fixed ratio thresholds. The image mosaic based on improved image registration:Using the improved SIFT algorithm extracts the features, and achieves the feature matching with the adaptive ratio thresholds. The result indicates that two methods achieved the seamless mosaic about images those translated, rotated with big angles, or scaled, but it can evidently increase the efficiency of image mosaic based on improved the image registration.