【作者】 潘喆；

【导师】 吴一全；

【作者基本信息】 南京航空航天大学 ， 信号与信息处理， 2010， 硕士

【摘要】 随着汽车拥有量急剧上涨,智能交通成为一个新兴的研究方向。图像处理技术是智能交通系统的重要前沿研究领域,具有十分重要的理论意义和应用价值。本文主要研究对车牌图像、车辆图像及道路图像的阈值分割技术。本文首先对智能交通技术作了概述,对图像阈值分割的一般方法进行综述,指出了现有算法存在的不足,相应地提出了新的算法。二维Shannon信息熵法是图像阈值分割中常用的经典算法,但存在着不足。为此,提出了二维指数信息熵阈值选取方法,克服了对数熵的不足,提高了速度。同时给出其快速算法,将二维阈值转化为一维。结果表明,该算法能快速准确地实现图像分割。最小类内方差法(Otsu法)分割精确,应用范围广泛,实质上是最小二乘法(L2范数)。本文提出了基于最小类内绝对差(L1范数)及最大差(L∞范数)的阈值分割算法,导出了这两种算法的二维算法。通过比较,发现这两种方法在某些类型图像下,阈值分割效果明显优于最小类内方差法,其二维算法的分割效果普遍优于相应的一维算法。基于熵的阈值选取方法是一类颇受关注的阈值分割算法。本文提出了二维Renyi熵的阈值分割两种不同的快速递推算法,都可以将计算复杂性由O(L4)减少为O(L2)。实验表明,二维Renyi熵阈值选取算法分割准确,这两种快速算法的运行时间不到原算法的0.1%。本文在指出现有二维直方图区域直分法中存在明显错分的基础上,提出二维直方图区域斜分法,导出了二维直方图斜分的Otsu阈值选取公式及快速递推算法。结果表明,二维直方图斜分可使分割后的图像内部区域均匀,边界形状准确,更具有稳健的抗噪性,运算速度大大提高。最后,本文将群智能中的人工鱼群算法应用到图像阈值分割算法中,提出了二维Otsu阈值分割的人工鱼群算法。通过与基于基本遗传算法及最优保存策略遗传算法进行比较,发现人工鱼群算法能够准确找到最佳阈值,收敛速度更快。更多还原

【Abstract】 With the increase of vihecles, the intelligent traffic becomes to a new research direction. Applying image processing technology to intelligent traffic system is a challenging field which has great theoretical significance and practical value. This paper concerns on the thresholding segmentation on the images of licese plate, vihecle and road.Firstly, the intelligent traffic techinology is introduced. On the basis of reviewing the method of image thresholding segmemntation, we point out the shortages of the existing algorithms and propose several new algorithms.The two-dimensional Shannon’s information entropy is a classical and commonly used image segmentation method, but there are still some disadvantages involved. Then, a two-dimensional exponent information entropy for threshold selection is proposed here, which could overcome the disadvantages in the Shannon’s entropy. Meaanwhile, a fast algorithm of two-dimensional exponent entropy thresholding method is also given, which changed the two-dimensional threshold into one-dimensional. The results of the experiment indicate that the proposed algorithm has high speed of calculation and good segmentation quality.The Minimum Within-Cluster Variance algorithm (Otsu) has a good segmentation quality and wide suitable scope, which is actually the Least-Squares algorithm (L2-Norm). Two algorithm for thresholding are proposed in this paper, which are based on Minimum Within-Cluster Absolute Difference (L1-Norm) and Minimum Within-Cluster Maximum Difference (L∞-Norm). And the corresponding two-dimensional algorithms of those two new methods are also presented. The results show that those two new methods have much better perfoemance for some kinds of images, and each of the two-dimensional algorithms is better than its own one-dimensional algorithm.Thresholding algorithm based on entropy is one of the most famous methods. In this paper, two fast recurring two-dimensional Renyi entropic thresholding algorithms, whose computational complexities are both only O(L2), are proposed, while the computational complexity of the original algorithm is O(L4). Experimental results show that these two recurring algorithms can both greatly reduce the processing time of images, which is less than 0.1% of the original algorithm.Baed on the obvious wrong segmentation in the existing two-dimensional histogram vertical segmentation method, a two-dimensional histogram oblique segmentation method is proposed. Then the formula and its fast recursive algorithm of the maximum Shannon entropy thresholding are deduced based on the two-dimensional histogram oblique segmentation. Experimental results show that the proposed method makes the inner part uniform and the edge accurate in the threshold image, and it has a better anti-noise property with the increase of its speed.In this paper, the fish-swarm algorithm of swarm intelligence is also introduced in image thresholding, and the two-dimensional Otsu thresholding algorithm baes on the fish-swarm algorithm is proposed. Comparition with single genetic algorithm and the elitist strategy genetic algorithm shows that this algorithm could select the best threshold accurately with a faster convergent speed.更多还原