【作者】 张震；

【导师】 邹永魁； 马驷良；

【作者基本信息】 吉林大学 ， 计算数学， 2009， 博士

【摘要】 本文就数字图像处理中边缘提取和去噪技六进行了研究。首先,系统介绍了数字图像处理领域的主要研究内容,指出了数字图像的边缘提取和去噪对于后续处理工作的重要意义。接下来,回顾了从Fourier分析到小波分析以及在小波基础上发展起来的各种应用调和分析工具的发展历程,分析了它们各自特点和优缺点,着重介绍了本文所采用的方向小波交换的理论基础和优势。作为本文的主要研究工作,在边缘提取方面,我们提出了基于方向小波交换边缘提取算法和改进的基于Canny算子的边缘提取算法;在噪声图像恢复方面,提出了基于方向小波变换的高斯噪声图像去噪算法以及基于BP神经网络和开关中值滤波的脉冲噪声图像恢复方法。最后,简要总结了本文内容,对本文算法需要完善和继续深入研究的内容进行了说明,就本领域关心的一些问题提出了我们的想法。更多还原

【Abstract】 Digital image processing technology originated in the 1920s, up to early 1970s, has formed a more perfect system of discipline. With the constant deepening of the study, people come to realize the edges of the images contain most useful information which is the main media of the image information. It is significant to extract effectively the edges of the images for image segmentation, image recognition and so on. In the actual images, their quality sometimes degenerate seriously as a result of noises. It is no doubt that the studies about the mechanism of the images degeneration, noise characters, how to establish the degradation model as well as how to restore effectively the images will be benefit for the following edge detection work. This dissertation concentrates on the edge detection and the denoising about digital image processing technologies. In edge detection aspect, we proposed an edge detection algorithm based on the directional wavelet transformation and an improved edge detection algorithm based on the Canny operator; In the denoising aspect, a denoising algorithm about the Gauss noise image based on the directional wavelet transformation and a restoring image algorithm about impulse noise based on BP-net and adaptive window switching median filtering are proposed.(1) An Edge Detection Approach Based on Directional Wavelet Transform.We propose an edge detection approach based on directional wavelet transform which retains the separable filtering and the simplicity of compu- tations and filter design from the standard two-dimensional WT. This separable discrete directional wavelet transform is implemented based on lattice theory. Although the transforms can be applied along any directions, only four transform directions {[1,0], [0,1], [1,1], [-1,1]} are chosen because of considering computational complexity and correlation among pixels. In this new transform frame, the corresponding gradient magnitude is redefined. In the process of applying directional wavelet transforms along four directions {[1,0], [1,1], [0,1], [-1,1]}, the eight directions with respect to directional derivative are generated, namely {[1,0],[-1,0]; [1,1],[-1,-1]; [0,1],[0,-1]; [-1,1],[1,-1]}. We find that it is unsuitable to apply non-maximum suppression only along maximal change direction of the derivatives in a number of experiments. In some cases, it is still necessary to choose the secondary change direction of the derivatives. Based on the analysis mentioned above, an new algorithm for non-maximum suppression is described.(2) Improved Image Edge Extraction Algorithm Based on the Canny Operator.In the Canny edge detection algorithm, edges are detected and located by double threshold values, so that it is difficult to choose a reasonable Lower limit threshold value. Choosing a big lower limit threshold will lead to troubles that edges cannot be fully located and are discontinuous; On the contrary, choosing a small lower limit threshold will produce a number of false edges. In addition, the edges which are gained by the traditional Canny algorithm cannot achieve the single pixel level, that is, An edge point corresponds several responses. We have improved the Canny edge detection algorithm. Locate the edge by a method with four threshold values. Then, conduct edge thin operation by introducing into a morphological operator. The improved algorithm can enhance capability of suppressing noise, delete false edges and obtain exact edges. Experimental results have indicated the feasibility and validity of the improved algorithm.(3) A Denoising Algorithm For The Gauss Noise Model Images Base On The Directional Wavelets Transform.At present, wavelet denoising has become one of the main image denoising methods. Based on some understandings and summaries for the current wavelet de-noising literature, we introduce the threshold de-noising and the relevance de-noising methods based on the standard wavelet respectively. Considering that the directional wavelet transform has more ability of directional transform than the traditional wavelet transform and the standard wavelet transform can be seen as a special case of the directional wavelet transform, we popularize the thoughts from the threshold de-noising and the relevance de-noising methods based on the standard wavelet into the directional wavelet. Our scheme is described as followed: Applying wavelet transform along several directional combinations, and then, letting median value of all groups as the final result, which can be considered as processing a rotating image. It is benefit to eliminate so called Gibbs affects and protect the image edges.(4)A Restoring Impulse Noise Image Algorithm Based On BP-Net And Adaptive Window Switching Median Filtering.In order to denoise impulse noises in images, an adaptive window switching median filtering method is proposed. It is different from the traditional median filtering method that our algorithm profits the thoughts from [84, 85, 86, 87, 88, 89, 90, 91], adopt two step schemes including noise detecting and filtering to denoise. Since only noise pixels are filtered, it can be avoided that the restoration image are seriously degenerated. In the stage of the noise detection, an noise detection method based on BP-Net is proposed. Firstly, the BDND algorithm from [92] is improved, which acts as a weak classifier and applies an initial classification for each pixel; and then, it is made a final decision by adopting the BP-Net. The method based on two steps makes the accuracy of the noise detector is improved significantly, which can detect effectively the distribution of the impulse noise in 70% noise density. In the stage of filtering, considering the traits of the noise detection method, a new adaptive window switching median filtering method is proposed. According to result of noise detection, the filter can adjust adaptively window’s width and sample choicely, each noisy point in image is denoised by filtering. Two benefits are significant. Firstly, merely signal point involve in filtering treatment, which avoid the interference from the noise points during filtering process; secondly, the filter can adjust adaptively window’s width, which can avoid the window is too large or too small to cause image blurring and distortion, protect effectively the edges and details of the images.更多还原