Single Image Super-resolution Reconstruction Based On Support Vector Regression

Super-resolution(SR)image reconstruction,as one of digital image processing methods,is widely used for safety monitoring,multimedia technology,remote sensing satellite and medical imaging,etc.And high resolution(HR)images based on fixed hardware can be obtained by using it.It's a process which produces a high resolution image from one picture or several low resolution(LR)pictures.In this paper,two different methods which combined support vector regression(SVR)and raster scan actions were proposed.They were all focused on single frame image and advanced the experiment results than some methods.In the end,an algorithm evaluating platform using hardware was presented,namely OpenCV.The main works include:1)A single image SR reconstruction approach using SVR and raster scan actions were proposed in discrete cosine transform(DCT)domain.And SR pictures were obtained by learning the relationship about some HR images and their LR editions.Compared with the bicubic interpolation method,the proposed method improves its Peak Signal-to-Noise Ration(PSNR)and Structural Similarity Index Measurement(SSIM)by 1.7% ~ 5.5% and 0.7% ~ 10% on different images,respectively.2)Another SR method was presented in spatial domain.It made full use of SVR and raster scan actions as information was extracted from only one image.And models which had learned were optimized.It enhances its PSNR and SSIM by 3.1% ~ 5.3% and 1.5% ~ 8.1% on different images,respectively as compared with the bicubic interpolation method.The final reconstruction effect was still very good.3)Hardware implementation of reconstruction methods based on OpenCV which has a good transplantable character was achieved.