【作者】 侯燕；

【导师】 祝宇鸿；

【作者基本信息】 吉林大学 ， 通信与信息系统， 2008， 硕士

【摘要】 随着多媒体的应用日益广泛。图像,已日益深入到人们的生活中。图像在采集、传输、解码过程中都有可能产生噪声。噪声使图像出现颜色的失真,清晰度下降,影响图像质量。由于产生的原因不同,噪声有多种类型,其中脉冲噪声最为常见,对图像质量有很大的影响。本文针对彩色图像脉冲噪声的去除算法进行了研究。根据彩色图像的特性,选择了结合了矢量中值滤波器(VMF)和矢量方向滤波器(VDF)的优点的方向-距离滤波器(DDF)作为基础。并提出了两种改进的噪声去除算法。第一种算法首先进行了区域检测,然后使用第一种噪声检测算子,最后使用高斯加权窗的DDF算法,处理被噪声污染的点;第二种算法是在第一种算法的基础上改进的,它的特点是运算量比第一种改进算法小,而且使用第二种检测算子,将疑为噪声点分为孤立噪声点和细节区,并利用VMF和DDF的优点,对上述区域分别使用VMF和高斯加权窗的DDF算法。仿真结果表明,使用了第一种改进的噪声去除算法的处理之后,图像的噪声去除效果较好,而且细节和色度保持很好。而且,使用了高斯加权窗的DDF方法比经典的VMF和VDF有更好的细节保持效果和较好的色度保持效果;而第二种改进算法,相比第一种改进算法是一种快速检测算法,目的是为了在图像中区分孤立噪声点和细节区,并利用VMF和DDF各自的优点。实验结果证明,第二种改进算法也具有较好的噪声去除效果。更多还原

【Abstract】 The color image processing system has been employed in many fields, from image-capture, to feature-abstraction, and the most common processing technique is filtering. And for the most cases, the noises are additive. Filtering is of the most importance. While during the filtering techniques, the most common one is the linear filter, and it is well-known for the simple mathematic expression, the uniform theory basics and simple implementation. However, most of them are on the base of Gauss statistics model.But, many questions could not been solved by the linear filtering technology. For example, the linear technologies have been limited by the application of image processing. Traditional linear technologies could not become the model of image-forming or image-transportation, and it has not considered the nonlinear model of human vision system. The images compose of smooth fields and the detailed fields, especially, the details mean the better to vision system. The filters which have the characteristic of detail-preservation have gained more and more applications. But most of the linear processing systems will make the details and the borderlines vague.The noises will degrade the comprehension of human to images, so it becomes the limited factor. Similarly, if you put the images to the quantitative analysis, the noises will restrict the performance of the systems. As a result, noises compression has been the key part of the image processing system, which restores the original images from the images polluted by the impulsive noises.There are noises in the process of images forming, transportation, and the decoding. The noises make the images color distortion, definition degradation. Due to the different reasons of generation, there are several types of noises. And the most generic one is the impulsive noise, and its impact to images is not negligent. Transportation errors, random movements of camera, instability of electrons, electromagnetic disturbance, errors of sensors, it is common in the transportation in the airs, including the light, the mechanical movements, the motor engines, or high-voltage line errors, and all kinds of electronic switches. And these become the interferences to the broadcast and the communication signals, while the corresponding images will be noisy. But the huge errors will generate a kind of impulsive noises, which appears as black or white pot, so it is called Peppers and Salt noises.The median filter is well-known for its effects of noise suppression to impulsive noise, but stretches the median filter to the color images simply, that is to imply the median to the RGB channel separately, will not generate good effects, since there are relations between the three channels. So the color distortion will appear. Due to these reasons, the vector median filters appear.Because the shortcomings of general vector algorithms are: they consider the threads and details as noises, then to process as usual. The image quality will decline, too. Therefore, noise-detection algorithm is to identify the noises, fine details, at the same time, separate the marginal fields and the details from the points which were noise polluted. That is the aim of noise-detection. But the most important thing is to distinguish between lines or the details and the noisy points, so, we introduced the edge-detection algorithms, thereby, to suppress the noises. This paper summarizes the main representative algorithms of several edge- detection algorithms and noise-detection algorithms, firstly, introduced some common edge-detection algorithm, and then introduced several representative algorithms, the first is CANNY edge-detection operator, which is a kind of gradient algorithm operator, the second one is the CIELAB color space noise-detection algorithm.To the gray images, the edges could be seen as the points whose gray scale jump, or they can be described as the gradual change, or the direction of the maximum gradient, in the gray-scale images, gradient operator can be used to detect image edges, but it is effective to the step edge only. When the gray width increases, the second derivative is a good method, instead, for example, the Laplace operator. And the edges could be described as follows: the zero crossing point of the second derivative. However, in the color image processing, color images are of the multi-channel, so vector processing method is more applicable.Therefore, in this paper, two types improved noise detection operator have been proposed. The first type is based on the vector computing method, and the second kind is a rapid detection, after the detection, the images were dealt with different methods. Firstly, VMF and DDF filter have been chosen as a basis, but in order to choose a common filter window, so the simulation analysis has been implemented. In this paper, improved color image filtering algorithms have been divided into two types: the first type algorithm, the image of the first divided into different districts, then imply the first noises detection operator to detect the noises points, finally, to used the improved filtering window to the DDF, or VMF; and for the second type algorithm, it is the fast detection algorithm, and it deal with the noisy point and the details with different methods. In the first type algorithm, it is on the basis of the first improved noise-detection algorithms, because it has compared the pros and cons of VMF and DDF, therefore, in the second type of algorithm, firstly, to use the rapid detection, which detected the noisy points, at the same time to avoid unnecessary treatment, such as the threads, then the noisy points could be detected and divided into the isolated noisy points and the details fields, and finally treat these noisy points with different algorithms, for example, VMF and DDF.Vector algorithm, in the applications of multi-channel color image processing, has occupied a very important position. However, based on different actual requirements, different algorithms are needed to different fields. For the first category algorithm, it simply choose the common window, and then detect the noisy point, the results have proved that, the result is better, its objective indicators have improved significantly, but this improved algorithm has the different the effect to the different images, which have different characteristics. And the phenomenons of noisy point-leakage exist, which is a flaw of this algorithm. The second improved algorithm is improved algorithm on the base of the first algorithm that is the analysis of the first algorithms. But in the future, the applications of the vector algorithms yet to be developed, but the effect is not very obvious, this also needs to continue to improve in the future.The simulation results show that the use of the improved noise-detection operator after processing, the effect of image noise removal improved, and maintain a very good details and color preservation. At the same time it could be seen that, the difference of three Operators of the first type, is not great and the choice could be according to the practical needs. Moreover, the use of the Gaussian window in the DDF or VMF, has gained the better effects than the classical VMF and DDF, and has the better property of details preservation and color preservation. And the second operator, in order to use their respective advantages of VMF and DDF, distinguish between noise in the image area and details.In short, because the images can be easily polluted by the noises of the wireless systems and the noises have the serious impact on the quality of the image. Therefore, removal of the noise in the image to the image processing is a very important part, and, noise suppressing, to the feature-extraction, target- identification, region-segmentation is a very important premise. Noise-suppressing in the video, is also on the base of image processing.Vector color image filtering is a new development area. Image noise-suppressing and correction is of very important significance, and they have now become emerging in the field, such as medical images, geological images, cultural heritage reservations, video communications, image post-processing machines, objective- identification, and so on. For the further, the vector image filtering technologies have opened a world of limitless possibilities.Color image processing technology in the present image science, communications, and multimedia applications has a very important position.更多还原