基于自适应非局部稀疏编码图像去噪方法

Adaptive nonlocal sparse coding for image denoising

由于图像的降质属性,传统的稀疏表示方法并不能如实的重建原图像。为提升基于稀疏编码方法图像去噪能力,提出一种非局部自适应稀疏编码图像去噪算法。为改进稀疏水平以及图像的局部属性,提出一种自适应学习字典;图像的非局部自相关先验融入到正则项中,提出一种自适应非局部正则项,进一步提升图像的去噪能力;为提高算法的有效性,利用一种迭代阀值算法进行优化。实验结果表明,该方法相对于BM3D、EPLL等方法具有较高的峰值信噪比(peak signal to noise ratio,PSNR)和结构相似度(feature similarity,FSIM),在图像细节、边缘保持和抑制视觉块效应方面具有比较好的重建效果,具有广泛的实际应用价值。

Due to the degradation property of the observed image, traditional sparse coding models may not be accurate enough for a faithful representation of the original image. To improve the performance of sparse coding-based image denoising, a simple yet effective framework for adaptive nonlocal sparse coding for image denoising was proposed. With the enhancement of sparse level and image local features, an adaptive learning dictionary was proposed. Meanwhile, the image nonlocal self-similarity was integrated into regularization term, and an adaptive regularization term was proposed to further improve the quality of image denoising. To enhance the computational efficiency of the proposed method, a fast implementation using iterative shrinkage thresholding method technique was developed. Experimental results demonstrate that the proposed method can effectively reconstruct the fine structures and edge preserve, and suppress the visual artifacts, outperforming many current state-of-the-art methods （i. e. , BM3D, EPLL） in terms of PSNR and FSIM and extensive practical application values.