High-resolution Image Restoration Based On Optical Imaging Characteristics

Even if optical image is almost principal information carrier in our life, with the development of technology, we wish more details from it. Especially for remote sensing, high resolution and clarity image is of great importance for meteorology, agriculture, forestry, natural resources exploration, and national defense particularly. However, the resolution of real imaging system is always limited by several factors. For example, information out of optical diffraction limit frequency is lost, while large filed aberration results space-variant degradation. Moreover, for detection system on the aircraft, vessel or vehicle, image is always with defocus and relative motion blur, global or local. To reconstruct the lost information, the paper discusses the image high-resolution restoration technology based on optical imaging characters.Firstly, we simulate the diffraction degradation and restoration for smaller relative aperture imaging system. Then a smaller aperture imaging system laboratory model for remote sensing application is set up. After capturing the images with different apertures, we compare the different restoration of them.Secondly, for large filed imaging system with space-variant point spread function (SVPSF), we suggest a total variation (TV) sectioned restoration algorithm based on Gradient Ringing Metric (GRM) image assessment. And a weighting splicing method is proposed to eliminate the ringing artifacts between image sub-sections. In addition, the relationship between the size of overlapping-region and the image quality of sectioned restoration is discussed. Then we complete the reconstruction of SVPSF image caused by SVPSF imaging system aberration.At last, we propose an object-segmentation based sectioned restoration for local blur image. A combination of several blur detection features is used as the "trimap" for image matting, which reduce the manual intervention. And the alpha map calculated by image matting serves to eliminate the interference of different areas in the weighting deconvolution process. The remained gap between blur and distinct areas is filled with the help of image inpainting algorithm.