Super-resolution image reconstruction based on MWSVR estimation
Hui Cheng, Junbo Liu · 2008
Super-resolution image reconstruction has been one of the most active research areas in recent years. Based on the theory of statistical learning, Mercer condition and the wavelet frame, this paper proposes a new multiscale wavelet support vector regression model (MWSVR) to reconstruction super-resolution image from low-resolution image and missing data image. The SVM essence is kernel method and the different kernel function has decided the different SVM. The choice of kernel parameters also is crucial in SVR function estimation. The MWSVR improve kernel function, and then the choice of kernel parameters is simplified in MWSVR, so the proposed model has wider applying scope. By the experiment with the single-variable two-variable function and real image, the new model not only can approach linear and the non-linear combination functions very well, but also performs better in Super-resolution image reconstruction. The results indicate that the proposed method has considerable effectiveness in terms of both objective measurements and visual evaluation.