基于纹理指示图的非局域去噪改进算法

非局域均值算法（NLM）是一种有效的去噪算法，能够在去噪的同时充分保护图像纹理信息。但是在像素的平滑中，权重分配过度依赖像素块相似性，缺乏其他更有效信息的指导，另外原算法计算量过大也是制约其应用的一个瓶颈。提出了一种基于图像纹理指示图的NLM快速改进算法TNLM。该指示图利用二维DCT变换对图像每一个像素为中心的小块进行分析，获得全图像素的纹理重要性指示图，将该指示图引入到传统NLM算法以改进去噪效果。同时，也对算法进行加速优化，通过改进的积分图技术、快速傅里叶变换和块预选技术，提高了算法的速度。实验证明，该算法的处理效果和速度都优于传统算法。

Image texture indicator diagram based improved non local means algorithm

Non Local Means（NLM） is an effective algorithm in image denoising, which can preserve the edge and textures in procedure of removing noise. However, in the smoothing of local noisy pixels, the assignment of the pixels’ weight depends too much on the patches’ similarity, without any other useful guidance. Another deficiency of NLM is the large computational time. This paper proposes a texture indicator diagram based improved NLM algorithm—TNLM. It calculates this diagram with the help of two-dimensional Discrete Cosine Transform（DCT）, which is performed over local patches in the whole image. This diagram is then adopted by the proposed method to improve the original NLM. Furthermore, it also cares about the acceleration of the algorithm, with the help of Improved Summed Square Image（ISSI）, Fast Fourier Transform（FFT） and patch pre-classification. The experiments also invalidate the effectiveness and efficiency of the proposed important map based on the analysis of each pixel patch. It incorporates this map into the NLM algorithm to improve the weights calculation strategy. Furthermore, the proposed method also accelerates the original method. The experiments validate the effectiveness and efficiency of the proposed TNLM.