Image Mosaic And Cylindrical Panorama Optimization Based On SURF

Vision is one of the important ways for human beings to experience the information of the world and image is one of the main expressions of visual information. With the advent of information era, the demand of image information is higher and higher. While the ability of image collection is limited to the hardware device, so how to capture high quality image at a low cost with digital image process becomes a hot research field. Image mosaic technology is developed in such a background, with which people can get a higher-resolution image with a lower cost. Image mosaic can mainly be divided into three parts: image pre-process, image matching and image fusion. This dissertation is mainly concerned with the core and the related theory and algorithm of image mosaic. The author’s major contributions are outlined as follows:1. In image stitching, well performed stitching result is hard to achieve for images with large difference of contrast. According to image pre-process, this paper proposes a method based on histogram equalization by enhancing the contrast of stitching images. On the basis of image enhancement this method uses the algorithm of SURF(Speeded Up Robust Feature) feature points, K-nearest neighbors and bilateral matching method to match the feature points. Further, the method uses RANSAC(RANdom SAmple Consensus) algorithm to eliminate the miss matched points and LM(Levenberg–Marquardt) algorithm to generate homography. Experimental results show that the method of the paper can realize image stitching with large difference of contrast.2. SURF has reached sub-pixel level in matching accuracy and is much faster than SIFT(Scale-invariant Feature Transform), but in the whole process of image process the part of feature points extraction is still the most time-consuming. According to this, for image feature extraction algorithm this paper proposes a SURF feature detection algorithm based on overlapping areas. Considering the particularity of image mosaic, ignoring the rotation of image and determining the overlap ratio of images this method only extract the feature points in the overlap region and uses these points to generate homography. Experimental results show that the method not only keeps the robustness of SURF but also greatly shorten the time of feature extraction and generation of homography.3. The purpose of image fusion is to stitch discrete images to a complete image on the vision and without stitching seams. During the image fusion, if there exists moving objects in the images to be stitched the traditional linear weighting method may generate ghost in the final image. According to this, this paper proposed a new method called four-section linear weighting method. Experimental results show that the method of the paper can eliminate ghost phenomenon to a certain extent.4. By using the theories and algorithms mentioned above, the cylindrical panoramic image mosaic is realized in the paper and the improvement of efficiency by using the improved methods in cylindrical panorama is analyzed. The paper also proposed a variable-slope straightening method to straighten the output panorama with wave effect and a simplified brightness corrective method to uniform the brightness of the final panorama.