Fusion-based effective noise removal approach with the reconstruction of guided reference image

Abstract

Image denoising approach has been studied for many decades. The main focus of image denosing is how to preserve image detail while remove image noise, however, it is hard to precisely distinguish the detail from noise. Up to now, even the state-of-the-art methods have the disadvantage of smoothing the detail with the noise at the same time. Inspired by the guided image filter(GIF) approach, we come up with a brand new approach to eliminate the image noise without losing the detail. A high quality guidance image is reconstructed by the specific fusion method of the near infrared image and the RGB image. With the reconstructed guidance image, the GIF approach can provide precise guidance filtering effect on the noisy RGB image. This approach performs good under strong noise level without smoothing the detail. Theoretical analysis and experimental results demonstrate that our method performs much better than the proposed methods. Examples have given illustrations of the proposed method.