光学学报, 2013, 33 (4): 0411001, 网络出版: 2013-03-05
基于稀疏重构的空间邻近目标红外单帧图像超分辨方法
Super-Resolution Method of Closely Spaced Objects Based on Sparse Reconstruction Using Single Frame Infrared Data
成像系统 稀疏表示 超完备字典 空间邻近目标 超分辨 红外图像 1范数正则化 imaging systems sparse representation overcomplete dictionary closely spaced objects super-resolution infrared image 1-norm regularization
摘要
针对现有算法难以利用单帧红外图像实现空间邻近目标(CSO)的超分辨问题,提出了一种基于稀疏重构理论的单帧超分辨方法。该方法充分利用了目标在焦平面阵列(FPA)分布的稀疏性以及光学系统点扩展函数(PSF)的结构特性,通过对FPA离散化网格采样构造稀疏量测模型,并将建立的1范数正则化问题转化为二阶锥规划问题求解;然后针对稀疏度过估计的重构结果,采用贝叶斯信息准则(BIC)实现模型选择,最终获得对目标个数和位置的准确估计。多组仿真场景验证了算法的有效性和超分辨能力;相比于已有算法,所提算法不仅提高了分辨正确率和位置估计精度,同时大幅缩减了计算耗时。
Abstract
Since the state-of-the-art methods barely have the capability of super-resolving the closely spaced objects (CSOs) using only single frame data, a super-resolution method based on the sparse reconstruction technique is proposed. The proposed method combines the sparsity of the distribution of CSOs on the focal plane array (FPA) and the structure characteristic of the point spread function (PSF) to construct a sparsely represented measurement model by discretizing the image plane with sampling grids. Then the 1-norm regularization problem is efficiently solved by a second order cone programming framework. For the overestimated sparsity after reconstruction, the Bayesian information criterion (BIC) is utilized for the model selection. The estimated number and positions of CSOs are precisely ascertained at last. Several scenes are set to inspect the efficiency and the super-resolution capability of the proposed method. It indicates that the sparse reconstruction-based method outperforms the existing methods in the ratio of correct detection, the precision of position estimation and the computation load.
张慧, 徐晖, 林两魁. 基于稀疏重构的空间邻近目标红外单帧图像超分辨方法[J]. 光学学报, 2013, 33(4): 0411001. Zhang Hui, Xu Hui, Lin Liangkui. Super-Resolution Method of Closely Spaced Objects Based on Sparse Reconstruction Using Single Frame Infrared Data[J]. Acta Optica Sinica, 2013, 33(4): 0411001.