Super resolution is a technique that obtains high-resolution (HR) images from the corresponding low-resolution (LR) ones and moreover, makes these HR images as natural as possible. There are usually some artifacts of these HR images caused by the lack of the information of the detailed textures (high-frequency information). In our research, an algorithm based on weighted vector quantization is proposed to predict the high-frequency information. First of all, the classified vector quantization is adopted to establish a pair of codebooks of high-frequency information for LR and HR patches. Second, three code vectors are selected from the LR codebooks based on the correlation coefficients to establish a multilinear regression model with the input LR patch and meanwhile the weights are decided adaptively. Then, three corresponding HR code vectors are used with the weights to reconstruct the high-frequency information of the HR patch. Finally, the reconstructed high-frequency information is used to refine the preliminarily upscaled HR image. The experimental results show that the performance of the proposed algorithm is good not only in objective but also in subjective measurement.
Superresolution (SR) algorithms have recently become a hot research topic. The main purpose of image upscaling is to obtain high-resolution images from low-resolution ones, and these upscaled images should look like they had been taken with a camera having a resolution the same as the upscaled images, and at least present natural textures. In general, some SR algorithms preserve clear edges but blur the textures, while others preserve detailed textures but cause some obvious artifacts along edges. The proposed SR algorithm presents the detailed textures and, meanwhile, refines the strong edges and avoids causing obvious artifacts. The goal is achieved by using orthogonal fractal as the preliminary upscaling method in conjunction with the proper postprocessing where directional enhancement is adopted. In fact, the postprocessing part in the proposed SR algorithm can effectively reduce most jagged artifacts caused by SR algorithms. In the simulation results, it is shown that the proposed SR algorithm performs well in both objective and subjective measurements. Moreover, most detailed textures are properly enhanced and most jagged artifacts caused by SR algorithms can also be effectively reduced.
A good superresolution (SR) algorithm obtains high-resolution (HR) images from the corresponding low-resolution (LR) ones and, moreover, makes the former look like they had been acquired with a sensor having the expected resolution or at least as “natural” as possible. In general, fast SR algorithms usually result in more ill artifacts in the enlarged image, while the well-performed ones usually have great complexity and take much more computing time. For this purpose, four efficient SR algorithms based on regression models are proposed. In the proposed SR algorithms, the difference of a natural HR image and an HR image obtained by fast interpolation is taken as the lost detail and is supposed to be composed of several different oriented details. By the self-similarity of the input LR image and its corresponding HR image, a regression model is established by the input LR image to decide the proper respective weights of these oriented details which is then used to reconstruct the lost detail of the natural HR image. As shown in the experimental results, the proposed SR algorithms not only perform well in both objective criteria and visual quality but also take less computing time than some well-performing algorithms.