Abstract
Almost image interpolation method in existence assume that image is contaminated with low noise level. In practice, on the low light situation such as outdoor night scene, indoor scene, image is corrupted with high noise level in imaging process. On this situation, first image is denoised, then is interpolated. After denoising process, the structure and texture property of image will change so that plenty of assumptions of interpolation method is not robust, so it bring about artifacts such as jaggies, blurring, and ringing, on the interpolated image. For this reason, we propose joint image denoising and interpolation under the nonlocal weighting estimation framework. For noisy image, assemble similar blocks and stack 3D blocks via nonlocal block matching on the denoising process, and then propagate the 3D blocks for more accurate missing HR pixel estimation on interpolation process. Compared with the progressively separate process of denoising and interpolation. Because of more correct local structure estimation, the interpolated image preserve edge structure well and cause less defects.