【作者】 谢雨来；

【导师】 李醒飞；

【作者基本信息】 天津大学 ， 测试计量技术及仪器， 2010， 硕士

【摘要】 目前,水下图像拼接技术存在实时性欠缺、鲁棒性不高的问题,极大地影响了水下视觉技术的发展。本文提出一种水下图像拼接方法,目的是对水下图像进行实时、鲁棒地镶嵌,以获取大范围、无缝隙的水下拼合图像。首先加入针对水下图像的预处理环节;然后对鲁棒性好但处理速度慢的SIFT算法进行改进,将基于SURF算法的图像配准方法应用于水下图像配准;最后,使用基于分水岭变换与最大流/最小截方法的搜寻最佳缝合线的水下图像融合算法实现图像融合,完成水下图像拼接。实验结果显示,水下图像预处理方法可以有效的改善水下图像质量,并为后续的图像配准做好准备;基于SURF算法的水下图像配准算法,在提升图像配准速度方面效果明显,具有很好的实时性,并且在图像配准的准确度方面与基于SIFT算法的方法相当,同时具有较好的鲁棒性;基于分水岭变换与最大流/最小截方法的搜寻最佳缝合线的水下图像快速融合算法,可以有效地消除融合鬼影,消除或减弱拼接缝隙,改善拼接质量。本方法满足对水下图像拼接技术实时性、鲁棒性和自动性的需要,为海下航行器完成特定的海洋探索、海洋监测任务,特别是在水下视觉导航方面提供一定的帮助,体现工程实用性。更多还原

【Abstract】 Currently, underwater images mosaic technology has problems of the lack of real-time performance and low robustness which greatly influenced the development of underwater vision techniques. This paper presented an underwater image mosaic method, which aims to in real-time and robustly, complete underwater images mosaic to obtain the large-scale, seamless mosaic images underwater. First, an underwater image preprocessing method was added. Then, SIFT algorithm, which has good robustness, but slow processing speed, was improved. Images registration method based on SURF algorithm was applied to underwater images registration. Finally, optimal seam finding underwater images fusion method based on watershed transformation and max-flow/min-cut was used to achieve underwater images fusion and complete underwater mosaic. The experimental results showed that underwater image preprocessing method can effectively improve image quality and be prepared for following underwater images registration. Underwater image registration method based on SURF algorithm which can obviously speed up image registration has good real-time performance and considerable robustness. And compared with SIFT-based image registration method, their accuracy was almost equal. Optimal seam finding underwater images fusion method based on watershed transformation and max-flow/min-cut can effectively eliminate ghosting, clear or weaken stitching seam, improving images mosaic quality.The presented method of underwater image mosaic method meets the needs of real-time performance, robustness, and automation, which can help underwater vehicles to complete specific marine inspection and ocean exploration tasks, especially in underwater visual navigation, reflecting engineering practicability.更多还原