Super Resolution Images Reconstruction Technique

Due to the cost and limitation of the physical device and all kinds of imagingenvironment factors, image resolution becomes lower. However, high resolutionimages are generally required in many applications. Super-resolution reconstructionuses multiple degraded images which are blurring, distorted, noise polluted to obtain ahigh resolution image. Its purpose is to make up the shortage of hardware andimprove the spatial resolution of the image from the software. High resolution imageshave been widely used in medical imaging, video monitoring, remote sensing, digitalTV and image compression, etc.In this paper, super-resolution reconstruction methods for the multiple imagesand single image are studied. Projection onto convex sets (POCS) method and theregularization method are studied for multiple images reconstruction. In addition,super-resolution method based on compressive sensing is studied for single imagereconstruction. The contributions and innovation of this paper mainly involve thethree following aspects:1. Projection onto convex set (POCS) algorithm to inhibit Halo is proposed forquality problem of the POCS super-resolution image reconstruction. In order toimprove the blur edge caused by bilinear interpolation, the wavelet bi-cubicinterpolation is proposed to obtain the initial image estimate of reconstructionalgorithm based on POCS. The point spread function (PSF) with edge-preservingproperty is proposed to reduce the amount of edge Halo in reconstruction result. Thelength-variable relaxation projection parameter, obtained by the correlation betweenlow resolution observation frames and reference frame, is adopted to reduce motionestimate error. Experiments show that the method in this paper effectively reduces theamount of edge halo and improves the quality of reconstructed high resolution image.2. For the problem of the reconstruction error caused by noise in regularizationmethod, I put forward an adaptive regularized super-resolution method based on noiseestimation. A rapid and effective noise estimation based on block is proposed. Itaccurately estimates the existing noise energy of the high resolution image in the eachiteration so that reduces the reconstruction error caused by noise. The noise energyconstraint is used to select regularization parameter, which makes the regularizationparameters adaptively adjust with the change of the noise. Experimental results show that the method in the paper effectively reduces the noise introduced by thereconstruction error, maintains rich edge information and obtains high quality image.3. For the problem that the compressed sensing reconstruction algorithm relieson the known sparse degree，the sparse adaptive subspace pursuit reconstructionmethod is put forward. It gradually increases the number of candidate atoms andmakes the sparse degree adaptively change. The discrete wavelet transform as a sparsedictionary is adopted to obtain the sparse representation of high resolution image.Gaussian random matrix is used to implement low-dimensional space mapping andobtain the low resolution images. Experiments show that the method in this papergreatly increases the reconstruction speed and effectively improves the quality of thereconstructed high resolution image.