【作者】 刘冬梅；

【导师】 刘贵喜；

【作者基本信息】 西安电子科技大学 ， 控制理论与控制工程， 2008， 硕士

【摘要】 图像拼接(image mosaic)技术是将一组相互间重叠部分的图像序列进行空间匹配对准,经重采样合成后形成一幅包含各图像序列信息的宽视角场景的、完整的、高清晰的新图像的技术。图像拼接在摄影测量学、计算机视觉、遥感图像处理、医学图像分析、计算机图形学等领域有着广泛的应用价值。一般来说,图像拼接的过程由图像获取,图像配准,图像合成三步骤组成,其中图像配准是整个图像拼接的基础。本文研究了两种图像配准算法:基于特征和基于变换域的图像配准算法。在基于特征的配准算法的基础上,提出一种稳健的基于特征点的配准算法。首先改进Harris角点检测算法,有效提高所提取特征点的速度和精度。然后利用相似测度NCC(normalized cross correlation——归一化互相关),通过用双向最大相关系数匹配的方法提取出初始特征点对,用随机采样法RANSAC(Random Sample Consensus)剔除伪特征点对,实现特征点对的精确匹配。最后用正确的特征点匹配对实现图像的配准。本文提出的算法适应性较强,在重复性纹理、旋转角度比较大等较难自动匹配场合下仍可以准确实现图像配准,实验证明该算法在配准方面具有良好的效果,算法准确率高,鲁棒性强,具有较高的使用价值。基于变换域的算法是图像配准常使用的算法,它思路简单、具有一定程度的抵抗噪声,但很难处理镜头存在旋转和缩放的情况,同时搜索整个图像空间,计算代价高昂。本文研究了基于傅氏变换的图像配准方法,引入的极坐标变换方法可以将旋转角度、比例缩放因子转换为平移的形式,大大减少了计算量。实验结果表明了此方法的可行性和有效性。如果没有拼合处理,那么我们得到的拼接图像将会有明显的缝隙,所以必须寻找一个良好的合成算法以消除这种缝隙。本文提出了一种改进的渐入渐出的拼合算法对配准后图像合成,得到无缝的拼接图像,取得了满意的拼接效果。更多还原

【Abstract】 Image mosaic is a technology that carries on the spatial matching to a series of image which are overlapped with each other, and finally builds a seamless and high quality image which has high resolution and big eyeshot. Image mosaic has widely applications in the fields of photogrammetry, computer vision, remote sensing image processing, medical image analysis, computer graphic and so on.Generally speaking, the image mosaic process consists of the following steps. image acquisition, image registration, image fusion. Image registration is the important foundation of image mosaic.This article has studied two kind of image registration algorithm. feature-based image registration and FFT- based image registration algorithm.Through researching the feature-based method, a robust image registration method is proposed. Firstly, corners are extracted using improved Harris operator which improves the precision and speed. Then, a normalized cross correlation and a kind of match method with bidirectional greatest correlative coefficient are used to extract the initial feature point pairs, and then the false feature point pairs are rejected by a Random Sample Consensus algorithm.Finally, the correct matching feature point pairs are used to realize the image registration. It has better registration results under a variety of conditions such as different light, bigger rotation and repetitive texture. The experiment results indicate this algorithm has good effect in image registration, high accurate rate, strong robustness, higher use value.FFT-based phase correlation image registration algorithm is often used. Its theory is simple, realizes easily, resists noise, but time is long and difficult to process the rotation and zooming. This article has studied FFT-based image registration method, the polar coordinate transformation is introduced in order to make the rotation and zooming factor transform into the translation form, greatly reduce the computation. The experimental result has indicated this method feasibility and validity.If without image fusion technology, the mosaic image will have the obvious seam, therefore must seek a good fusion method to eliminate this kind of seam. This article has proposed one kind of improvement gradually enters gradually leaves fusion method to realize the image mosaic. The algorithm has obtained the satisfactory blending effect.更多还原