基于K-LT的高分辨率图像的分块算法

Block Algorithms of High-Resolution Image Based on Karhunen-Loeve Transform

针对卡洛变换(Karhunen-Loeve Transform,K-LT)应用于高分辨率图像处理中,存在计算量大和速度慢的缺点,提出大分块算法和小分块算法以快速实现K-LT。大分块算法通过把图像矩阵均匀分块,得到多个分辨率相同的子图像,再把这些子图像纵方向堆叠形成伪多光谱图像来降低特征空间的维数;小分块算法则把每个子图像像素采用行堆叠或列堆叠的方法来降低特征空间的维数。2种算法都能够使K-LT速度大幅度提升。仿真结果表明:对于分辨率为1 024×1 024的图像矩阵,采用这两种分块K-LT算法所用的时间,分别是传统K-LT算法的1/45和1/48,可以满足实时性处理的要求。

In order to overcome the shortcomings of large amount of calculation and slow speed as the Karhunen-Loeve transform is applied to high-resolution image, this paper puts forward respectively a larger block and a smaller block algorithms to fast implement the Karhunen-Loeve transform. To reduce the number of dimension of the feature space and to improve the speed of the Karhunen-Loeve transform, in the first algorithm, the image is divided into sub-images with the same resolution, and then these sub-images are stacked vertically to form pseudo muhispectral images, while in the second algorithm, each sub-image pixel is stacked by row or column for the same purposes. For the image matrix with resolution of 1 024 x 1 024, the experimental results show that the execution time of two algo rithms respectively is 1/45 and 1/48 of the conventional K-LT＇s ,which can meet the real-time needs.