【作者】 王刚；

【导师】 朱宁波；

【作者基本信息】 湖南大学 ， 计算机科学与技术， 2010， 硕士

【摘要】 图像分割作为图像处理、分析和理解等领域的一项关键技术,其作用是把图像分成各具特性的区域并提取出感兴趣的目标。二值化是一种直接、有效的图像分割方法。通过二值化,可以将一幅图像分割为各自灰度值相似的两个区域,从而可以直观的呈现出图像中的目标和背景。由于图像通常受各种未知因素干扰,二值化算法较难分割出准确的目标和背景,同时算法复杂度也相对增高,制约了其在实际工程中的应用。因此,研究快速、高效的二值化方法在图像分析、理解、目标识别等技术中具有重要的意义。本文着力于研究快速、有效的二值化算法,主要工作有：1)针对一维Otsu算法抗噪性差和二维Otsu算法复杂度高的缺点,提出一种基于直方图投影的二维Otsu快速算法。通过将二维统计直方图投影到其对角线,形成峰谷分布明显的一维统计直方图,利用一维Otsu算法得到对角线上的分割点,并在基于此分割点和对角线垂直的直线上搜索最佳阈值,使二维阈值搜索转换为两次一维搜索,提高了算法搜索速率。同时,通过建立基于二维阈值(s,t)的查找表,加快了算法的计算速度。实验证明,该算法具有较好的抗噪性和实时性,不仅保留了二维算法的分割效果,且分割速度与一维算法相当。2)针对Gato算法在处理低质量文档图像时依赖于其他局部二值化算法的缺点,提出一种改进的基于目标估计和GRA增强的二值化算法。改进的算法通过使用修正的均值滤波器估计图像背景,并使用GRA增强算法对目标图像进行增强,最后用快速二维Otsu算法进行二值化。实验证明,该算法对于受阴影覆盖、低对比度等低质量文档图像具有较好的分割效果。3)针对某企业的实际工程需求,设计和实现了一个视频监控识别系统。该系统在监控视频中截取图像,通过进行一系列处理后对图像内容进行识别。在该系统的预处理及识别过程中,应用了本文提出的二值化算法进行目标分割,算法能够满足系统对功能和速度的要求。目前该系统已投入使用。更多还原

【Abstract】 Image segmentation is a key technology in image processing, analysis and understanding areas. Image segmentation technology segments image into regions each with characteristics and extracts interesting targets form image. Binarization is the most commonly used method of image segmentation. By binarization, an image can be divided into two regions, each with similar gray value. By this way, the foreground and background could be presented visually. However, in most of time, we can not get correct result of foreground and background after binarization because of various unknown factors in images. Meanwhile, the complexity of binarization methods is becoming relatively higher, which limited its applications in practical engineering. Therefore, studying on fast and efficient binarization methods in image analysis, understanding, object recognition and other technology has important significance. In this paper, we focus on studying fast and efficient binarization algorithm. The principal works are as following:1) For the shortcomings of poor performance in noise immunity of one-dimensional Otsu algorithm and alarming calculation of two-dimensional Otsu algorithm. We proposed a fast two-dimensional Otsu algorithm based on histogram projection. We projected the two-dimensional histogram onto its diagonal, which formed a one-dimensional histogram with marked distribution of peaks. Based on the one-dimensional histogram, we got a point on the diagonal by one-dimensional Otsu. Then we searched the optimal thresholds on the line which is perpending to the diagonal by two-dimensional Otsu algorithm. By this way, we changed the two-dimensional threshold search into twice one-dimensional search with the same effectiveness of image binarization. Furthermore, we established three look-up table based on the two-dimensional threshold (s, t), which could accelerate the computation speed. Experiments showed that the algorithm had better noise immunity and real-time. The algorithm not only retains the segmentation of two-dimensional algorithm, but also has the segmentation speed fairly with one-dimensional algorithm.2) Gato’s binarization algorithm depends on other local binarization methods in processing the degraded document images. To overcome this week point, we proposed an algorithm based on foreground estimation and GRA enhancement. In our new algorithm, we estimated background image with a modified mean filter. Furthermore, we enhanced the foreground image by GRA method. Finally, the fast two-dimensional Otsu algorithm was utilized to get binarization result. Experiment results showed that our method performed well in degraded document images.3) We designed and implemented a video surveillance recognition system according to an enterprise’s work. The system intercepts images from monitored video, and recognizes the content from images after a series of processing. We made use of our new binarization method in image pre-processing and recognition of the system. The new algorithm can satisfy the system’s requirements in functionality and speed. The system has been put into use presently.更多还原