基于稀疏表示的图像复原算法研究

发布时间：2019-06-21 10:47

【摘要】：在图像获得的过程中,由于成像条件和外界环境的干扰,往往会使图像质量下降,但是在实际生活中,又需要高清晰的图像。因此,有必要从退化的图像中恢复出高清晰,高质量的图像,这就是图像复原的首要任务。另一方面,图像复原作为一种底层图像处理技术,在恢复出高质量图像的同时,也给后续图像处理奠定了质量基础,因此图像复原已成为图像处理研究领域关注的基本问题。图像复原是指应用图像的先验信息,建立合适的模型,从退化的图像中恢复和重建原始图像的一种图像处理技术。它主要包括图像去噪,图像去模糊,图像修复和提高分辨率几个方面的问题。本文主要围绕图像去噪和图像去模糊两个方面做了深入研究。论文首先对图像复原的研究背景与意义,图像退化模型,图像复原的一般方法进行了简单的分析与总结。然后,基于全变分正则化在图像复原的过程中会产生阶梯效应这一现象,本文将图像的稀疏先验信息引入图像复原问题中,设计了一种新的图像复原模型。在本文设计的图像模型中,由于紧小波框架可以有效的对逐段光滑函数进行稀疏表示和自适应获得当前待处理图像的多尺度边缘结构,本文设计的模型可以有效地克服全变分正则化带来的阶梯效应。此外,由于本文模型的不光滑性,我们采用增广拉格朗日算法进行数值求解。实验结果证明,针对高斯噪声情况下的图像模糊问题,本文设计的模型可以有效复原图像,并消除阶梯效应,且复原效果优于现有的图像复原方法。一般情况下,研究学者主要考虑了受高斯噪声污染的图像去模糊问题,但是在图像成像的过程中,往往也会受到脉冲噪声的污染,因此,我们考虑了受脉冲噪声污染的图像去模糊问题。由于脉冲噪声的成像机理与高斯噪声成像机理不一样,所以针对高斯噪声污染的图像去模糊问题的方法不能直接用于受脉冲噪声污染的图像去模糊问题中,鉴于此,本文结合紧小波框架和全变分正则化设计了一种新的图像复原模型,并采用增广拉格朗日算法对其求解。实验结果证明,本文设计的方法可以有效地处理脉冲噪声污染的图像去模糊问题。最后基于自适应字典的稀疏表示方法可以有效地去除高斯噪声,但是却不能很好地去除脉冲噪声。因此,针对受脉冲噪声污染的图像去噪问题,本文在自适应字典稀疏表示的基础上设计了一种二阶段脉冲噪声去除方法。首先利用中值类型滤波器将图像分为噪声点和非噪声点,然后建立基于l1-l1最小化的字典学习方法,并采用交替方向方法进行数值求解。实验结果证明,本文提出的方法在有效去除噪声的同时可以很好地保存图像信息。
[Abstract]:In the process of image acquisition, due to the interference of the imaging conditions and the external environment, the image quality is often reduced, but high-definition images are required in the real life. Therefore, it is necessary to recover high-definition and high-quality images from the degraded image, which is the primary task of image restoration. On the other hand, image restoration, as a bottom-layer image processing technique, provides a quality base for subsequent image processing while restoring high-quality images, so that image restoration has become a basic problem in the field of image processing research. Image restoration is an image processing technique for restoring and reconstructing an original image from a degraded image by using the prior information of the image, establishing a suitable model, and recovering and reconstructing the original image from the degraded image. It mainly includes image de-noising, image de-blurring, image restoration and resolution. This paper mainly studies the two aspects of image de-noising and image de-blurring. In this paper, the background and significance of image restoration, the image degradation model and the general method of image restoration are analyzed in this paper. In this paper, we design a new image complex by introducing the sparse prior information of the image into the image restoration problem based on the phenomenon of the step effect in the process of image restoration based on the full-variation regularization. In the image model designed in this paper, because the compact wavelet frame can effectively perform the sparse representation of the piecewise smooth function and the multi-scale edge structure of the current image to be processed, the model designed in this paper can effectively overcome the order of the full-variational regularization. Ladder effect. In addition, due to the inhomogeneity of the model, we use the augmented Lagrangian algorithm to count. The experimental results show that the model designed in this paper can effectively recover the image and eliminate the step effect, and the recovery effect is better than that of the existing image. In general, the research scholars mainly take into account the image de-blurring problem of the Gaussian noise pollution, but in the process of image forming, the pollution of the impulse noise is often also affected, so we take into account the image of the pulse noise pollution As the imaging mechanism of the impulse noise is different from the Gaussian noise imaging mechanism, the method of image de-blurring for the Gaussian noise pollution cannot be directly used for the image de-blurring problem of the pulse noise pollution. In this paper, in view of this, a new image restoration model is designed by combining the compact wavelet frame and the full-variational regularization, and the augmented Lagrangian calculation is used. The experimental results show that the method designed in this paper can effectively deal with the image of pulse noise pollution. Finally, based on the sparse representation of the adaptive dictionary, the Gaussian noise can be effectively removed, but it can't be very good. To remove the impulse noise, a two-phase pulse is designed on the basis of the sparse representation of the adaptive dictionary for the image de-noising problem of the impulse noise pollution. the method comprises the following steps of: firstly, dividing an image into a noise point and a non-noise point by using a median type filter, and then establishing a dictionary learning method based on the l1-l1 minimization, The results of the experiment show that the method proposed in this paper can be very good at the same time.
【学位授予单位】：武汉大学
【学位级别】：博士
【学位授予年份】：2014
【分类号】：TP391.41

【参考文献】