Font Size: a A A

Color Image Super Resolution Reconstruction Based On YUV Model In Wavelet-Domain

Posted on:2009-07-19Degree:MasterType:Thesis
Country:ChinaCandidate:J SunFull Text:PDF
GTID:2178360242476701Subject:Control theory and control engineering
Abstract/Summary:PDF Full Text Request
In many applications, recorded images represent a degraded version of the original scene because of the limitation of some physical conditions. And these low-resolution images can not be satisfying practically. So to enhance the resolution of the recorded images using super-resolution reconstruction algorithm is attracting more and more attention. Image super-resolution reconstruction refers to the technology of estimating or reconstructing a (or a sequence of) high-resolution undistorted image(s) from a (or a sequence of) low-resolution distorted image(s) while eliminating the additive noise and the blur brought by lineate detector and optical element.The super-resolution reconstruction is mainly comprised of the super-resolution reconstruction of a single image, the super-resolution reconstruction of a sequence of static images, the super-resolution reconstruction of video sequences and the super-resolution reconstruction of compressed streaming videos. And only the first one, i.e. there's only one low-resolution image, is discussed in this paper. The super-resolution reconstruction comes down to such difficulties as pixel registration, identification of blurs, effective reconstruction algorithm and so on.In these days, most of researches, in the field of image super-resolution reconstruction, are based on grey-scale images. However, it's very significant to obtain high-resolution color images in the field of remote sensing, outer space exploration, high definition television, medical imaging, etc. An algorithm based on HMT (Hidden Markov Tree) model in wavelet-domain is put forward by Zhao Shubin, Zhang Peng,[1] etc. and it has gained a higher signal-noise ratio and a good visual effect. But the shortcoming is that it contains model parameters training, multiple iterative computations and adaptive image transform due to using the RGB model of a color image, so the time complexity is very high. Therefore, this paper proposes an algorithm on the basis of wavelet-domain local Gaussian model and YUV model of a color image. In this algorithm, the luminance channel (Y channel) is reconstructed using wavelet-domain local Gaussian model super-resolution reconstruction and the other two chromatic aberration channels (U,V) are reconstructed using a lower-resolution (relative to Y channel) reconstruction. In this case, computation time could be greatly improved while ensuring good visual effect.In this paper, the first part introduces the basic concepts and mathematical description of image super-resolution reconstruction and some classical methods in details such as IBP (Iterative Back Projection), POCS (Project Onto Convex Sets), MAP (Maximum A Posteriori), MLE (Maximum Likelihood Estimation) and so on. And then it extends to the color images. At the end of this part, it presents the measurement standards of the quality of reconstructed image(s).The second part of the content focuses on the local Gaussian model in wavelet domain. Firstly, it introduces the regularization, and describes the image super-resolution reconstruction in wavelet domain. Secondly, it dissects local Gaussian model detailedly based on which as the prior probability to regularize the reconstruction in wavelet domain. Lastly, to realize the super-resolution reconstruction with the Y, U and V channels reconstruction combined together.In the end, doing simulation according the algorithm proposed in this paper and analyzing the experimental result. It's demonstrated that the reconstructed high-resolution image using this algorithm not only has a high PSNR (Peak signal noise ration) and a good visual effect but also reduces the computational complexity significantly.
Keywords/Search Tags:super resolution, YUV model, wavelet transform, local Gaussian model, color image
PDF Full Text Request
Related items