基于各向异性的遥感影像匀光匀色算法

为解决现有的遥感影像匀光匀色方法关注重点停留在全局、半全局统计参数的变换基础上，存在全局性的特点，缺乏局部特征信息引导的内容驱动，无法顾及到每张影像内容的差异性，导致匀色效果不佳，影像对比度损失，残留色差明显等问题，本文通过分析影像间出现色差的本质原因，把同名特征在不同观测环境下的观测值作为观测条件，并辅助加以局部空间约束与改正量正则约束，研究构建针对具有重叠区影像集的各向异性优化方法，利用凸优化理论求解出特征点处的改正量，然后以此改正量作为锚点空间辐射至整幅影像，线性加权改正得到最优匀光匀色效果。结果表明本文提出的基于各向异性的遥感影像匀光匀色算法，相对于传统的匀光匀色算法，理论跟完备，无假设条件，效果更佳，重叠区域影像局部范围内色差更小。可见本文提出的匀光匀色算法具有效果的有效性，和对各向异性匀光匀色算法研究的参考价值。

Research on light and color uniformity algorithm of remote sensing image based on anisotropy

The existing methods of uniform light and color for remote sensing images are based on the transformation of global and semi-global statistical parameters. They lack the content drive guided by local feature information and cannot take into account the difference in each image content, resulting in poor color uniformity, loss of contrast, and obvious residual color difference. In this paper, by analyzing the essential causes of color difference between images, the observation values of homonymous features in images under different lighting environments were taken as the observation conditions, and local space constraints and correction regularization constraints were supplemented to study and construct an anisotropic optimization method for image sets with overlapping areas. The method used convex optimization theory to solve the correction amounts at the homonymous feature points, and then used the correction amounts as the anchor points to radiate the space to the whole image and obtained the optimal uniform light and color effects through linear weighted correction. The results have showed that, compared with the traditional algorithm, the proposed algorithm for light and color equalization of remote sensing images based on anisotropy was more complete in theory, without assumptions and preconditions, have achieved better results, and with less chromatic aberration in the local range of the overlapping area images. It can be seen that the uniform light and color algorithm proposed in this paper is effective and progressiveness, and this research has reference value for the research of anisotropic uniform light and color algorithms.