具有敏感区域的华南虎图像配准研究

对数极坐标映射是一种具有刚体不变性的模板匹配算法,以它获得的刚体变换矩阵为初值经LMA(Levenberg Marquardt algorithm)迭代可以实现图像的高精度配准,但对于场景发生剧烈变化的图像对,上述方法无法获得正确的配准参数。为此,提出了一种具有敏感区域的图像配准算法,在LMA迭代中仅考虑落在人工选取的敏感区域中的像素点,通过对陕西林业厅公布的华南虎照片进行的图像配准实验表明了该算法的有效性。

Image Registration of the South China Tiger with Sensitive Region Concerned

The Log-polar mapping is a rigid invariant image matching algorithm,Levenberg Marquardt Algorithm(LMA)can calculate image registration parameters with high precision,but the LMA algorithm fails when facing an image pair with intense scene variance.In this paper we propose an image registration algorithm with sensitive region introduced in LMA phase,only pixels in the manually selected sensitive region are calculated in the iteration.The algorithm is used to register five representative photos of the Shanxi South China Tiger.The experiment shows that the method is a valuable improvement to the LMA algorithm.