【作者】 康牧；

【导师】 王宝树；

【作者基本信息】 西安电子科技大学 ， 计算机应用技术， 2009， 博士

【摘要】 近年来,随着图像应用领域的拓宽,图像处理技术得到了迅猛的发展,已经成为图像理解和计算机视觉领域中一项重要而有用的技术。同时,传统的基于空域和频率域的图像处理方法都存在这样或那样的不足,提高这些图像处理算法的效果和运算速度也就显得特别有意义。图像处理就是采用一定的算法,对一幅质量不高的图像进行处理,使处理后的图像具有更高的清晰度、更好的可理解性。它不仅可以使处理后的图像更适合人的视觉观察,而且可以为进一步的图像应用提供更有效的信息,例如图像的分类、图像分割、目标识别与目标检测等、战损评估与理解等。现有的图像处理方法主要有两类:一类是空域中的处理,即在图像空间域中对图像进行各种处理;另一类是把空间域图像经过变换,如傅立叶变换,变换到频率域,在频率域中进行各种处理,然后再变换回图像的空间域,形成处理后的图像。这两类方法都很常用,但具体到其中的每一种方法,则各有长短,它们大部分是适合于某一类图像的处理,并且在该类图像处理方面能够在一定程度上提高性能。本文对图像处理中的图像去噪、图像增强、边缘检测、边缘细化、图像缩放等算法进行了分析,提出了相应的改进算法,所做的主要工作有:1、结合人眼视觉特性提出了一种新的计算阈值的方法,该方法是根据人眼在不同灰度区域具有不同的敏感度提出来的,充分考虑了人眼视觉的灰度特性,由于它是基于人眼的视觉特性提出来的,所以可以适用于大多数可见光图像,该方法计算的阈值应用于图像去噪、图像增强和边缘检测等方面都取得了比较好的结果。2、提出了一种具有去噪功能的图像增强算法,该算法结合人眼视觉的灰度特性和结构化特性,定义了新的梯度算子,该算法不但可以增强图像的视觉效果,还可以有效地抑制图像的噪声、增强图像的细节信息,抑制噪声的效果比较理想,另外,该算法不仅可以用来增强灰度图像,也可以用来增强彩色图像,且不会引起图像失真。3、说明了RGB模型在图像增强中的合理性,指出了HSI模型在图像增强方面的不足。4、针对Robert边缘检测算法对噪声比较敏感,提出了一种改进的Robert边缘检测算法,该算法不但具有一定的抑制噪声的能力,检测到的边缘也更细腻光滑;针对Kirsch、Prewitt边缘检测算法的复杂度比较高,结合有关学者的思想对它们进行了分析和改进,提高了它们抑制噪声的能力,对于Prewitt算法还降低了它的复杂度。针对传统边缘细化算法复杂度比较高,分析了它的原理,简化了它的算法,在不影响运算结果的前提下,降低了算法的复杂度。5、根据提出的弹性模型设计了图像放大和缩小算法,实验结果比较理想。6、根据物理学中的惯性原理进行边缘跟踪,分析和改进了Canny边缘检测算法,在没有提高算法复杂度的前提下,提高了Canny边缘检测算法抑制噪声的能力,且检测到的边缘更加细腻光滑。7、将经济学中的贫富差距原理应用到边缘检测中,分析和改进了Robinson边缘检测算法,降低了算法的复杂度,进一步提高了抑制噪声的能力。文中所提出的每种算法,图像边缘检测、图像增强和图像缩放,均给出了相应的实验结果,并得到了一些有价值的结论。通过对实验结果以及对算法主、客观评价分析表明,文中给出的算法是有效的。更多还原

【Abstract】 In recent years, the image processing technology has been developed quickly with the widening of image application fields. The image processing has been an important and useful technique in the fields of image understanding and computer vision. In the same time, traditional image processing methods based on space domain or frequency domain exist some shortcomings, which is very meaningful to improve the effect and speed of these image processing methods. The image processing is meant to use some algorithms to process a poor-quality image, which the purpose is that the processed image has higher definition and more understandability. It can not only be suitable for human to observe but also provide useful information for further applications, for example, image classification, image segmentation, object recognition, object detection, battle damage evaluation and so on.Now the image processing methods can be classed into two types. One method is image processing based on space domain, which the image is processed in space domain. The other is transforming the image from space domain to frequency domain, such as Fourier transformation. The transformed image is processed in frequency domain. Then the processed image is transformed to space domain again. These two methods are all useful, but each of them has advantages and disadvantages. In general, one method is only suitable for a certain class of images and only can improve the performance in a certain extent for the class of images.In this dissertation, the methods of image restraining noise, image enhancement, edge detection, edge thin and image zoom are all investigated and the improved methods are proposed. The main works are done in this dissertation have:1. A new method for calculating threshold according the property of people’s vision is proposed. Based on the different sensitivity of people’s vision at the different grey scale area, this method is proposed, which the property of people’s vision grey scale is sufficiently considered. Since it is proposed according the property of people’s vision, so it can be suitable for lots of visible light image. The threshold calculated by this method can obtain good effectiveness in image restrain noise, image enhancement and edge detection.2. According the property of people’s vision grey scale and structure, an image enhancement algorithm with restrain noise function is proposed, which defines a new template to calculate gradient. This algorithm can not only enhance the vision effect of image, but also can restrain noise and enhance the detail signal. The effect of restraining noise is very good. On the other hand, the method can enhance not only the grey scale image, but also the color image, which can not distort the color.3. The rationality of model RGB in enhancing image is illustrated and the shortcoming of model HSI in enhancing image is proposed.4. For the sensitivity to noise of Robert edge detection algorithm, an improved Robert edge detection algorithm is proposed. It has the ability of restraining noise, and the detected edge is more smoother. For the complexity of Kirsch edge detection algorithm and Prewitt edge detection algorithm is comparatively high, they are analyzed and improved according the thoughts of some scholar. Improved their ability of restraining noise. At the same time, for the Prewitt algorithm the complexity of algorithm is declined too. Because the complexity of original edge-thin algorithm is high, its principle analyzed and its algorithm are simplified. The complexity of algorithm is declined under the condition of keeping the original effect.5. The algorithm of image zooming and image shrinking according the proposed model of spring is proposed. The results of experiment are very good.6. According to the principle of inertia in physics to trace the edge, Canny edge detection algorithm is analyzed and improved. The ability restraining noise is improved under the condition of keeping the original complexity, and the edge detected is smoother.7. Using the principle of disparity of between poor and rich of economics into the edge detection, Robinson edge detection algorithm is analyzed and improved, which the complexity of algorithm is declined. The ability restraining noise is improved more.All results of experiment are given for every algorithm of image edge detection, image enhancement and image zoom in the dissertation. Some worthy conclusions are also obtained through analyzing the experimental results and performance of objective evaluation the subjective analysis for the algorithms. All the experimental results show that the algorithms are effective.更多还原