【摘要】 提出了一种基于过完备空时字典及其稀疏表示的红外小弱目标运动检测算法。采用K奇异值分解算法学习连续多帧图像的运动信息和形态特征,构建自适应形态过完备空时字典;利用高斯运动模型检验自适应形态过完备空时字典,将其划分为能分别描述目标与背景的目标过完备空时字典和背景过完备空时字典;将连续多帧图像分别在目标过完备空时字典和背景过完备空时字典上稀疏分解,利用几个最大稀疏系数及其空时原子重构信号,增强二者残差来检测小目标信号。实验结果表明,该过完备空时字典不仅能同时描述目标与背景的运动信息和形态特征,极大地提高信号表示的稀疏程度,而且能有效增强目标与背景的特征差异,提高小运动目标的探测能力。更多还原

【Abstract】 A dim moving target detection algorithm based on over-complete spatial-temporal dictionary and sparse representation is proposed. A spatial-temporal adaptive morphological over-complete dictionary is trained and constructed according to infrared image sequence. It can represent the motion information and morphological feature of target and background clutter. The spatial-temporal morphological over-complete dictionary is subdivided into two categories: target over-complete spatial-temporal dictionary for describing moving target,and background over-complete spatial-temporal dictionary for embedding background. The criteria adopted to distinguish the target spatial-temporal redundant dictionary from the background spatial-temporal redundant dictionary is that the atom in target over-complete spatial-temporal dictionary could be decomposed more sparsely over Gaussian over-complete spatial-temporal dictionary. Subsequently,the image sequence is decomposed on the target and background over-complete spatial-temporal dictionaries,respectively. The dim moving target and background clutter can be sparsely decomposed on their corresponding over-complete spatial-temporal dictionary,yet it couldn’t be sparsely decomposedon their background over-complete spatial-temporal dictionary. Therefore,the target and background clutter would be reconstructed effectively by prescribed number of atoms with maximum sparse coefficients in their corresponding over-complete spatial-temporal dictionary,and their residuals would differ so visibly to distinguish target from background clutter. The results show that the proposed approach not only could improve the sparsity more efficiently for dim target image sequence,but also could improve the performance of small target detection.更多还原