关键词: 超分辨率重建, 正则化, 双边全变分, 自适应核回归

Abstract: Regularization method is used widely to solve the ill-posed problem in super-resolution reconstruction. Based on the analysis of the existing regularization super-resolution reconstruction algorithm, an adaptive kernel regression filtering core based on Bilateral Total Variation（BTV） is constructed, and it is used as a cost function of regularization super-resolution reconstruction. It can produce kernel function adaptive locally to image features. Compared with traditional regularized reconstruction method, the experimental results show that the algorithm can effectively remove noise, but also can be very good to retain the image details, and has robustness. The objective and subjective evaluation shows that the quality of reconstructed image has significantly improved.

Key words: super-resolution reconstruction, regularization, bilateral total variation, adaptive kernel regression