全景图像自动拼接算法的优化设计  

Optimized design of automatic panoramic images mosaic

作　　者：杨磊[1] 曹剑中[1] 唐利孬[1] 高博[1] 王华[1] 郭惠楠[1] 

机构地区：[1]中国科学院西安光学精密机械研究所,陕西西安710119

出　　处：《红外与激光工程》2014年第3期985-990,共6页Infrared and Laser Engineering

基　　金：国家自然科学基金(61201376)

摘　　要：提出了一种结合空域和频域进行全景图像序列自动拼接融合的优化算法。首先利用改进的相位相关法对全景序列图像进行自动排序并确定重叠区域,根据重叠区域像素均值进行图像整体亮度差异自动调整以降低角点的误匹配率;然后使用改进的Harris算子在空间域提取图像角点(无需人工设定阈值),通过双向最大互相关系数匹配获得初始特征点对,并用RANSAC算法实现精确匹配;最后采用非线性平滑算法对图像重叠区域进行融合处理。实验结果表明,该优化算法排序过程简单有效,特征点提取匹配过程的成功率和效率都较现有算法有很大提高,拼接的图像清晰度高,具有较高的稳健性和拼接精度。An optimized algorithm of automatic panoramic image mosaic was presented based on time and frequency domains. Firstly, the improved phase correlation method was used to sort the panoramic image sequence automatically and then determine the overlap region. In order to reduce the false match rate of the corner, the brightness of the overall image was adjusted automatically according to the overlap region＆#39;s pixel mean. Then, an improved Harris operator was adopted to extract corners （with no need to set threshold value manually）, the initial feature points pairs were obtained by the bidirectional greatest correlative coefficient and the false feature point pairs were rejected by RANSAC algorithm. Finally, the nonlinear smoothing algorithm was adopted to fuse the overlap regions of the image. The experimental results show that optimization algorithm is simple and effective in the sorting process. Compared with the existing algorithm, the optimization algorithm proposed at present makes a great progress in the success rate and the efficiency in the matching and extracting process of feature points. The stitched image is with great clarity, soundness and stitching accuracy.

关 键 词：图像拼接 相位相关 自动排序 角点检测 图像融合 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]