Study On High Resolution Computational Imaging With Coded Sensing

Technology of multi-dimension information acquisition is one of the tasks of thenational medium and long-term program for science and technology development. Highresolution imaging is an important research and exploration direction.Multi-dimensional high-resolution images are highly desired in many fields, such asspace remote sensing, medical diagnosis and military reconnaissance, etc.In the traditional imaging method, an image is acquired directly from the scene inthe spatial domain. There exists one-to-one relationship between the scene and theimage. The resolution of the image is equal to the number of sensors in the imager.Therefore, because of the limitation of imaging principle, spatial resolution and spectralresolution of the imaging system are restricted to the density and sensitivity of thespectral detector. It is difficult to obtain simultaneous multi-dimensional high-resolutionimages in the traditional imaging method. Many scholars have to face the challenge tofulfill high resolution imaging with low-resolution.As a rapidly developed technology in recent years, computational imaging, andcompressive sensing bring new hope for multi-dimensional high-resolution imaging.They have received extensive attention from scholars at home and abroad. Thisdissertation, under the support of “863” program, the national science foundation ofchina, fundamental research funds for the central universities of china, focuses onhigh-resolution computational imaging based on coded sensing, and mainly studies thecoded sensing methods of gray images, RGB full-color images and remote sensingspectral images.The main contributions and innovation points of the thesis are as follows:(1) Aiming at the problem of image spatial resolution restricted to the density of adetector, we propose a high-resolution imaging method via moving random exposure. Inour method, the image acquisition is performed in two stages, coded sensing(compressive measurement) and optimization reconstruction. In the first stage, alow-resolution camera along with the aircraft moves relative to the scene. compressivemeasurement are made by the low-resolution camera with randomly fluttering shutter,which can be viewed as a moving random exposure pattern. In the optimizationreconstruction stage, the HR image is computed by different models according to theprior knowledge of scenes. The simulation results demonstrate the effectiveness of theproposed imaging method. The proposed imaging method offers a new way of acquiring HR images of essentially static scenes with low-resolution, which is particularlysuitable for the occasions where the camera resolution is limited by severe constraintssuch as cost, battery capacity, size, memory space, transmission bandwidth, etc.(2) There exist some problems such as the low quality, staircase artifacts (zippereffects) in the traditional methods. This thesis reexamines the color demosaickingproblem in a perspective of sparsity-driven image restoration, and propose a new colordemosaicking with an image sparse model and adaptive PCA (Principal ComponentAnalysis). Our method describes the sparsity within each color component by imageformation model and spectral feature, excavates spatial sparsity by adaptive PCA. Thespectral sparse representation is derived from a physical image formation model; thespatial sparse representation is based on a windowed adaptive principal componentanalysis. So, our method exploits spectral and spatial sparse representations of naturalimages jointly, and further proposes an minimization technique for reconstruction.The simulation results demonstrate that our method outperforms many existingtechniques by a large margin in PSNR and achieves higher visual quality.(3) Aiming at the problems that spatial resolution and spectral resolution arerestricted to the density of the spectral detector and usually cannot be acquiredsimultaneously. This thesis, based on our coded sensing computational imagingframework, presents a high-resolution computational spectral imaging method ofremote sensing. The new image acquisition system employs a fluttering shuttercontrolled by a random sequence to modulate exposure to an ordinary imaging sensor,without change of mechanical structure and without increasing the density of originalimaging detector. It enables multiple scene pixel intensity accumulated in the samesensor pixel. Aliasing effect is produced. The optimization inversion algorithm based oncompressive sensing theory is used to reconstruct high-resolution multiple-spectralimage. Simulation results demonstrate that our method can greatly enhance the spatialresolution and keep high spectral resolution simultaneously without increasing densityof original imaging detector.(4) In the above spectral imaging method, the use of coded exposure shutter in anoptical system sacrifices the amount of exposure because a fluttering shutter blockslight, resulting in a poor SNR. So we further design a new coded exposure strategywithout reducing exposure and present a new code exposure method based on highspeed switching of reflection angle. In the proposed system, we install the randomlyrotating mirror behind the slit of the imager. The method uses the randomly rotatingmirror to divide the incident light into two beams and fulfills random exposure complementary modulation. All pixels in a scene participate in the aliasing andsufficient information is collected. The high spatial resolution multi-spectral image canbe recovered by exploiting the signal sparsity. The recovery algorithm is based oncompressive sensing theory. Simulation results demonstrate the efficacy of the proposedtechnique.