基于小波变换的图像处理方法研究（主要研究图像增强，包括源代码）

苏州科技学院本科生毕业设计（论文）

results,one can see that, both geometrical regularization methods and neighborhood filters have their own advantages in processing geometrical structures or textures of image.Through image feature decomposition, these advantages have been fused properly to form our proposed methods. This feature-dependent adaptive processing allows us to obtain better denoising and enhancement results.

The BM3D algorithm is used to denoise the noisy geometrical image.We an see that, although the BM3D algorithm preserves the sharpness of image edges and corners, it produces annoying ring effects around them, which are typical by-products of the filtering in the spectrum domain.From the 110th profiles of results,overshoots can be observed easily on edges compared with the anisotropic diffusion.Thus,for image edges a different processing is needed to adapt to these features.

2) Image sharpening:in Table 1 the PSNR results verify further the better performance of our proposed methods numerically.However,it is true that the BM3D algorithm often obtains the best PSNR result in image denoising.

The BM3D algorithm is used to denoise the noisy Barbara image,Although it has a better PSNR performance, it blurs image edges and textures compared with the AD+NLM algorithm. Thus, even if in a single denoising task we also need an image sharpening operation.

Considering comprehensively both the visual evaluations and the PSNR results, we think that the AD+NLM algorithm is a competitive method in image denoising and enhancement.

3) Some extensions: the proposed methods can be iterated recursively to recover further image information lost in the residual parts.A similar idea can be found. Moreover, one can use image decomposition methods to decompose an image into the structure part and the oscillating part,which are processed afterwards by the geometrical regularization method and the nonlocal means filter respectively,just as we did .Here, the focus of this paper is only simplicity and efficiency of the proposed algorithms. Thus, we will not discuss these cases.

46

苏州科技学院本科生毕业设计（论文）

Ⅴ.CONCLUSION AND FUTURE WORK

Adaptive processing of image features is very important in image denoising and enhancement,particularly for noisy textured image,This paper presents simple and efficient algorithms, which fuse advantages of geometrical regularization methods with better processing of image geometrical structures: edges,corners and details,and the nonlocal means filter with better denoising of image textures.At the same time,image sharpening is indispensable in image denoising.Results in numerical experiments show that, feature preserving or sharpening of image textures and fine details.

In the future work,we will develop single adaptive method integrating both smoothing and sharpening properly.

47