Research On Automatic Image Stitching Method For Unordered Images

Unordered image stitching is the process of combining multiple photographic unordered images with overlapping fields to produce a panorama or high-resolution image. The key technologies include image registration, image sequencing and image mosaic strategy and fusion. We focus on these four key technologies and propose a new automatic image stitching method for unordered images in this paper. The main work is as follows:1) A sparse representation bidirectional feature matching method based image registration method is proposedFirstly, SURF features are extracted from both images. Secondly, the features are matched by the features’sparsity. Thirdly, bidirectional identification method is used to gain the high precision matching result and the transform relationship between images is estimated. Finally, Lucas-Kanade algorithm is used to improve the accuracy of the registration method, which is initialized by the geometric transformation parameters computed from the matching points. The experimental results show that the registration algorithm can achieve a precision that higher than0.06pixels under effects of light or noise.2) A sparse representation bidirectional matching feature based image sequencing method for unordered images is proposedThe algorithm is based on the matching points got by the sparse representation based bidirectional feature matching method. The Abscissa of the matching points can determine the positional relationship between the two images. The matching degree between images is determined by the number of matching points. Image sequencing can be realized based on these two factors. The experimental results show that the algorithm is effective for actual shooting images.3) An image fusion method based on the local energy of wavelet decomposition is proposedFirstly the image is decomposed into wavelet coefficients. Secondly, the low frequency component is fused by " fade into and fade out" method. Thirdly, for the high frequency components with higher matching degree, they are fused by the weight of local energy; For the high frequency components with lower matching degree, the component with higher local energy is chosen. Finally, the fused image is obtained by wavelet reconstruction. The experimental results show that this algorithm achieves better result both objectively and subjectively.4) An image mosaic strategy based on passing multiplication style is proposedThe proposed image mosaic strategy is based on passing style and indirect style mosaic model. Firstly, adjacent two images are aligned into images with wider fields of view, and their vertex coordinates data are saved at the same time. Then the panorama image is got by aligned the images using the vertex coordinates data with passing style.The experiments on synthetic and actual shooting images show that the proposed method is insensitive to ordering, illumination changing and noise. It can automatic construct multiple panoramas from an unordered image dataset with higher definition.