Font Size: a A A

Research On Image Denoising And Fusion Algorithms Based On Multiscale Transform

Posted on:2014-02-13Degree:DoctorType:Dissertation
Country:ChinaCandidate:Y S LiuFull Text:PDF
GTID:1228330398959599Subject:Signal and Information Processing
Abstract/Summary:PDF Full Text Request
Multiscale transform is a sparse representation with limitations, directionality and multiscale characteristic. Multiscale transform can decompose the signal or image into different scales, and obtain the feature of the signal or image in different resolution, so we can get the accurate information of the signal or image. Because of the advantages of the multiscale transform, it has been applied in many areas, such as image processing, computer vision and pattern recognition. In the case of image processing, multiscale transform can choose the appropriate decomposition scale according to the actual needs, and it is helpful to get the needed image information, so multiscale transform is an important tool for image processing.Image denoising is a classic problem in image processing. When we get an image using a digital camera or traditional file cameras, the image may contain noises from different sources. Noise will affect the quality of the image, and it often has great influence on the subsequent image analysis. In order to get a reliable result, we need to remove the noise as much as possible and maintain the detail structure information such as edges and texture. Multiscale transform is one of the methods which can meet the above requirements of the image denoising, so it has been applied in image denoising domain extensively, and it is an indispensable means in the field of image denoising.The definition of image fusion is combining the images from different sensors for the same scene to achieve a more accurate and more specific interpretation of this scene. Image fusion technology has been widely used in many scientific research and applications, such as computer vision, remote sensing and military fields. According to the fusion hierarchy, image fusion can be divided into:pixel-level fusion, feature-level fusion and decision-level fusion. Pixel-level image fusion is the most common in image fusion, and multiscale transform is a widely used method in this field. It is the significant problem in image fusion that how to improve the integration of the final fused image resolution and keep the image information in the source images as far as possible.This thesis mainly does research on the theory and application of image denoising and image fusion. The main contributions and innovations are as follows: (1) A new image denoising method based on Dual-Tree Complex Wavelet Transform and adaptive windows is proposed.In the traditional image denoising based on wavelet transform, how to estimate the coefficient variance of the wavelet coefficients in different subband is a critical problem. Most paper chose the square neighborhood of the current point to estimate the coefficient variance. This form of neighborhood selection considers that the wavelet coefficients energy distribution of different direction is isotropic, but it does not match the actual distribution of the wavelet coefficients. An improved algorithm using ellipse windows as the local neighborhood is proposed. Six directional subbands can be obtained by one level of dual-tree complex wavelet decomposition, and dual-tree complex wavelet transform has approximate shift invariance. Those are proper characteristics for image denoising. In the different directional subbands of the same scale, the ellipse window is adopted to match the directions of the subband filters, and its size is determined according to the scale. The experiment results reveal that this method is simple and effective.(2) A method of despeckling of ultrasound images in contourlet domain is proposed.Recent years more and more public attention has focused on the accumulated dose in patients brought by the X-ray equipment. The radiation dose affects harmful in a very wide range. The ultrasonography can be part of the solution to such problems. Because of the particularity of ultrasound imaging mechanism there is a lot of speckle noise in the ultrasound image. The speckle noise reduces the contrast of the image and complicates the image processing tasks such as compression and segmentation. The speckle noise is considered as one of the major problems in the processing of the ultrasound images. Based on the property of contourlet transform and ultrasound images, we proposed a novel denoising method:a spatially adaptive threshold is introduced for denoising the contourlet coefficients of log-transformed ultrasound image. The threshold is obtained from the probability density function of relative variance of coefficients. The experiments results prove that our method is available.(3) A new image denoising algorithm based on Contourlet transform and anisotropic diffusion is presented. Contourlet transform is a classic method of multiresolution analysis. It can get different size and number of directional subbands on different scales, so it can better describe the geometric structure information and distinguish the noise and weak edges of the image. Partial Differential Equations as a powerful mathematical tool have been widely used in image denoising. The contrast factor of the diffusion function plays an important role of image denoising and edge preserving in the process of diffusion filter. It has been demonstrated that the correspondence between the diffusion function of anisotropic diffusion and the Tukey’s biweight error norm of robust estimation theory, so the robust scale of the image can be automatically estimated. After carrying on anisotropic diffusion to each directional subband of contourlet transform and inverse contourlet transform, the reconstruct image can be obtained.(4) A hybrid frequency image denoising method based on Gabor transform and Contourlet transform is proposed.The Gabor transform is a classic method of time-frequency analysis. Two-dimensional Gabor functions are widely used in image processing, computer vision, neuroscience and psychophysics. Gabor transform has been proved that it is the optimal description of signal spatial domain and frequency domain in the case of two-dimensional Uncertainty principle. Gabor transform is the’windowed’Fourier transform. It can effectively express the periodic and smooth texture part of the image. Contourlet transform is capable of retaining the structure information and the weak edges. Therefore the hybrid denoising method which combines the advantages of the two transform is presented, the wiener filter is used to filter the noise in two frequency domain. The experiment shows that the algorithm is effective.(5) A Contourlet-based multifocus image fusion algorithm using particle swarm optimization is proposed.Image fusion has become an important part of image processing. Multiple images can be taken from two or more sensors in the same time or different times for a specific image or scene. Image fusion is the process of combining the relevant information of these images to obtain a new explanation of the image or scene. A nonsubsampled contourlet transform (NSCT) based fusion algorithm with particle swarm optimization (PSO) is proposed. PSO is used to determine the parameter of fusion. The source images are decomposed using nonsubsampled contourlet transform. Depending on the nature of the transformed coefficients, they are fused by different fusion rules. Then the fused image is obtained by applying inverse nonsubsampled contourlet transform on the fused coefficients.
Keywords/Search Tags:image denoising, image fusion, multiscale geometric analysis, contourlet transform
PDF Full Text Request
Related items