Research On Temporal Suner-resolution Imaging Technology Based On Compressive Sensing

High time resolution is the ability to obtain optical images with high frame frequency.However,the charge readout rate of an imaging device,hardware restrictions on the transmission and storage rate of image data is the technical bottleneck to realize high frame rate photoelectric imaging.Meanwhile,the high frame frequency is formed a relation to influence and to condition each other with high spatial resolution,wide field of view and other imaging parameters,which makes it very difficult to obtain high resolution at the same time in multidimensional degree.Moreover,the constraints also lead to the failure of multi-observation to obtain sequence images in some situations.Therefore,high time resolution has a wide range of application requirements.The goal of research on temporal super-resolution imaging technology based on compressive sensing is to break the limit of device conditions.It obtains high time resolution images by reconstruction of observation data at a lower cost of data.That is obtain sequence images from one single image,which can effectively reduce the cost and difficulty of obtaining high time resolution images,and has great significance to reduce the amount of data storage and energy consumption.Furthermore,it also provides new ideas to meet high time resolution imaging requirements for high speed movement and rapid change of target observation,super-fast phenomenon and scientific exploration.It is of great scientific significance for the technological breakthrough and performance improvement of the photoelectric imaging system in the fields of scientific exploration,earth observation and surveillance and reconnaissance.Currently,most of the temporal compressed sensing reconstruction algorithms are very time-consuming,and the quality of the reconstructed image can be improved.This paper first studies the motion estimation and adaptive reconstruction of single image observation image.For the problem of the need of reconstructing the whole image,three adaptive temporal super-resolution imaging methods that only reconstruct motion regions are proposed,which are the motion area searching method based on speckle-like features,the motion estimation and reconstruction method based on signal correlation,and the motion estimation and reconstruction method based on interpolation.To avoid unnecessary reconstruction of a static background area,the first method using an equal length exposure coding mode to obtain a clear image with a static background region and then separate the motion area based on speckle-like features of these area by pixel similarity segment method.The separated areas are reconstructed and finally put back in the background image.Thus,the motion area can be separated from the single observation image without the need of reconstruction.So that,it greatly reduces the time consumption for reconstruction and the background noise reconstruction is avoided,which can slightly improve the quality of reconstructed images.In the second method,dictionaries are firstly trained from small pieces of video sample respectively according to the range of movements.Then the motion distribution information in the scene is estimated from a fast preliminary reconstructed image sequence based on the fact that the correlation coefficient of the image block decreases with the increase of the displacement.In the end,images are adaptively reconstructed by selecting the corresponding dictionary through the motion distribution map.Compared to the first method,this method can roughly estimate the distribution of the motion of the scene,and improve the quality reconstructed image in motion areas.In order to obtain more accurate estimation of motion distribution,in this paper,an interpolation-based estimation method of motion distribution is further proposed.Observation image is obtained by a specific continuous exposure mode,which can be interpolated to get high quality preliminary sequence images.By applying the first method that proposed in this paper,only the motion areas will be interpolated.Motion distribution map is then estimated from the interpolated images and used for adaptive reconstruction.This method can get the motion distribution more accurately from the single observation image.At the same time,there is no limitation on the reconstruction algorithm,which is suitable for most of the algorithms based on sample learning.Aiming at the problem that traditional blind restoration can only estimate the blur kernel but not get the motion information of the camera,in this paper,we combine the temporal super resolution technique with the image restoration technology and propose motion deblurring method based on local temporal compressive sensing.In this method,video blocks are reconstructed at the corners of the image sensor during a single exposure period.The displacement vector,which will be used to build the prior blur kernel for image deblurring,is then estimated from the reconstructed videos.With the use of the prior PSF,better recovered images can be obtained with much less iteration.This method can get the flutter information of the camera without additional devices and improve the quality of recovered images,especially for the degraded image with simultaneous defocus and motion blur.In the end,we also set up experimental system to verify the effectiveness of the proposed temporal super-resolution imaging methods and the restoration method.The experimental results show that the adaptive reconstruction methods based on motion distribution proposed in this paper can obtain the information of scene motion and improve the quality of reconstruction images.Meanwhile,the restoration method presented in this paper also has a good performance,which provides a new way of thinking for the traditional image processing.