多约束协同的彩色夜视影像亚像素超分辨率重建

Currently,super-resolution reconstruction methods based on image sequence are mainly for the higher quality images obtained from ideal imaging environment. Most of them only focus on the improvement of the spatial resolution. when these methods are used in night-vision images with low illumination, Movement blocked，higher noise, lack of color information, The effect is often very poor.. Consider the characteristics of night-vision imaging and human visual characteristics, night-vision multiple-constraints image super-resolution reconstruction under complex conditions are researched in these project. The ultimate goal is to improve theimage understanding and visual interpretation capabilities. Firstly, in order toobtain accurate motion information, high-precision low illumination TV-L1 (totalvariation -L1 norm) sub-pixel optical flow motion estimation method are established and improved.Secondly, building multi-scale bilateral filtering algorithm, which can keep theimage edge features well, to suppress noise amplification in the reconstructionprocess and adjust images high dynamic range. Then, in order to improve the colorinformation, the night-vision scene color reconstruction algorithm based on daylightimage statistical features and simplified color space is researched. Finally, the above-mentioned methods are integrated into the framework of POCS reconstruction,cooperated with each other, to greatly improve the image spatialresolution and color information.. This research has great significance for weak information processing in the night-vision image, such as segmentation, feature extraction, identification and interprets.

目前序列帧超分辨率重建主要针对理想成像环境中较高质量的影像，大多仅注重空间分辨率的提高，当应用于夜视影像（具有低照度、运动遮挡、大噪声、色彩信息不丰富等特点）时，效果往往不佳。. 本项目针对夜视影像特点和人眼视觉特性，将提高影像解译和可视判读能力作为最终目标，研究复杂夜视场景下的多约束夜视影像超分辨率重建方法。首先，为获取准确的运动信息，建立并改进低照度下高精度的TV-L1光流亚像素运动估计算法。其次，构建保持影像边缘特性的多尺度双边滤波算法来抑制重建中的噪声放大并调整影像的高动态范围。然后，研究基于日光影像统计特性和简化颜色空间的夜视场景色彩重构算法，改善色彩信息。最后，将上述多约束融入到POCS 重建的整体框架中，相互协同，迭代重建出空间分辨率和色彩分辨力都极大提高的影像。. 本研究对夜视影像中弱小信息的分割、提取、识别与解译等各种研究与应用都具有十分重要的意义。

本项目针对夜视影像特点（低照度、运动遮挡、大噪声、色彩信息不丰富等）和人眼视觉特性，研究并建立了复杂夜视场景下的多约束彩色夜视影像超分辨率重建方法。该方法在空间分辨率、色彩、纹理信息等各种细节上提高了夜视影像的质量，最终提高了夜视影像的可视判读能力。.在开展、完成项目的过程中，项目组按期提交了研究进展报告和结题报告。在项目的资助下，项目组已在高水平的国内外学术期刊/会议上发表论文13篇（其中SCI/EI 检索论文7篇）。以项目为依托，培养研究生11名，培养的研究生（已毕业）围绕项目研究已完成相关学位论文6篇。通过项目研究，目前已经建立了独立、稳定的科研团队。.本研究对于夜视影像中弱特征信息的处理，如弱小目标分割、特征提取、微弱信息识别与解译等各种研究与应用具有重要意义。

内容获取失败，请点击重试