摘要： 现有的视差图像拼接算法中单应性矩阵不具有全局性,且存在计算量大、拼接结果有重影和结构扭曲等问题。为此,提出一种新的的视差图像拼接算法。采用图割算法将参考图像和目标图像分割成若干个具有独特性质的图像块,并对图像块编号。运用SIFT算法对图像进行特征提取,在特征描述子中加入特征点的图像块信息,确定目标图像与参考图像之间的特征匹配图像块。通过特征分块法计算全局单应性矩阵,找出最优的全局单应性矩阵对目标图像进行预配准。在重叠区域加入形状扭曲约束和块链接约束局部优化预配准图像,得到修正后的图像,并进行图像融合获得全景拼接图像。实验结果表明,特征分块算法可确保单应性矩阵的全局性,减少迭代次数,提高计算效率,同时局部优化算法可以较好地消除重影和扭曲,保证图像拼接质量。

关键词: 特征分块, 特征提取, 特征匹配, 单应性矩阵, 局部优化, 图像拼接

Abstract: In the existing parallax image mosaic algorithm,homography matrix are not global,and there is a problem of large computation amount,overlapping ghosting and distortion of the structure.Therefore,a new parallax image mosaic algorithm is proposed.The reference image and the target image are divided into several image blocks with unique properties by using the graph cut algorithm,and the image blocks are numbered.The SIFT algorithm is used to extract the feature of the image,the feature block is added to the feature descriptor to determine the feature matching image block between the target image and the reference image.Through the feature block method,the global homography matrix is calculated and the optimal homography matrix is found to pre-align the target image.The pre-registration image is locally optimized by adding the shape distortion constraint and the block-link constraint in the overlap region to obtain a modified image,and the modified image is fused to obtain the panoramic mosaic image.Experimental results show that the feature block algorithm ensures the global homography,reduces the number of iterations and improves the computational efficiency,the local optimization algorithm can eliminate ghosting and distortion and ensure the quality of image mosaic.

Key words: feature blocking, feature extraction, feature matching, homography matrix, local optimization, image stitching

中图分类号: 

• TP391