基于全局和局部特征的图像拼接方法

针对传统序列图像拼接算法中的误差累积问题，提出一种基于全局和局部特征的图像拼接方法. 同时拍摄大视场角、低分辨率全局图像和小视场角、高分辨率局部图像，利用深度学习替代传统算法提取两者匹配点，进而根据两者面积比等比例扩大全局图像的匹配点坐标，将局部图像无缩放地投影至全局图像所在平面，最后融合投影后局部图像的重叠区域，拼接形成一幅大视场角、高分辨率全景图像. 实验结果表明，该方法中深度学习快速且精准地实现了特征匹配，同时局部图像间相互独立，有效地解决了拼接顺序限制和拼接误差累积. 

Image Stitching Method Based on Global and Local Features

To solve the error accumulation problem in the traditional sequential image stitching algorithm, a new image stitching method was proposed based on global and local features. Both one global image with large field of view and low resolution and some local images with small field of view and high resolution were taken simultaneously. Then, substituting deep learning for the traditional algorithm, the matching points of the two were extracted. And according to their area ratio, the matching point coordinates of the global image were scaled up at the same proportion for the purpose of projecting local images to the plane of the global image without scaling. Finally, the overlapping areas of local images after projection were fused and stitched to form a panoramic image with large field of view and high resolution. Experimental results show that deep learning can achieve feature matching quickly and accurately. Moreover, the local images are independent of each other, effectively solving the restriction of stitching sequence and the accumulation of stitching errors.