【作者】 董辛酉；

【导师】 曾培峰；

【作者基本信息】 东华大学 ， 计算机应用， 2008， 硕士

【摘要】 在纤维图像分析和自动识别系统中,需要用到近距离成像技术。但在实际的图像采集过程中,由于显微镜、成像系统、图像采集系统等各种环境的限制,采用简单的点光源,这种光源系统提供的照明是非均匀的,在图像的中心部分是直射,而在其它部分光线发生角度的偏差,使得图像上出现光照不均匀。非均匀光照在图像中产生背景噪声,它和图像信号混合在一起,造成图像的对比度和灰度分布不均。这种失真影响纤维识别处理算法的性能。所以,需通过修正算法消除光照不均引入的背景噪声。在有效地去除光照不均产生的背景噪声的同时,对图像中有用信息的保留是评价光照不均修正算法性能的一个重要特征。光照不均引起的背景噪声和图像中的有用信息在图像域以及频率域都存在重叠,因此必须讨论消除光照不均时有用分量的损失情况。本文首先对图像光照不均算法进行了分析。根据光照不均校正方法不同,划分为图像域和频率域两类方法,即基于图像域的照度不均校正技术和基于频域滤波的照度不均校正技术。前者对图像所在空间进行处理,图像域是指由像素组成的空间,图像域增强方法是指直接作用于像素的增强方法;后者对图像的处理是通过傅立叶变换、小波变换等变换方法,将图像转换到频率域通过分析图像信号的频率的不同进行处理操作。但各种不同的处理都有自己的不足之处。图像域方法为了得到比较好光照不均修正的效果,计算量较大;频率域方法由于滤波会在图像边缘产生模糊效应,而且滤波器会产生副作用。并且光照不均修正算法目前还缺乏统一的评价理论,这与没有衡量光照不均修正质量通用的、客观的标准有关。修正的方法往往具有针对性,修正后的结果只是靠人的主观感觉加以评价。为克服上述两种方法的缺点,本文提出了基于线性迭代的光照修正算法。首先,本文对图像信号的离均差进行线性迭代。在此迭代过程中,通过计算离均差偏离局部平均值的幅度构造特征点集合,并用特征点位置的局部强度平均值取代对应的强度值,不断消除强度值变化剧烈的边缘对强度值变化缓慢的边缘的影响。最终达到光照不均修正的效果。实验证明,线性迭代算法能够较好的去除光照不均背景噪声对图像的影响,能很好地解决图像边缘检测中光照噪声干扰和局部强度值对比失衡的问题,并且计算速度快。更多还原

【Abstract】 In the fiber image analysis and automatic recognition system, close imaging technology is required. But acquisition system uses spot lightening, that causes non-uniform illumination. The direction of lightening is upright in the center, but direction bias occurs in the other parts. When the image is contaminated by such interference, the image contrast and the gray level changes. That will affect the performance of the following algorithm in fiber recognition.It is expected that the non-uniform illumination should be alleviated and the useful information be preserved in the non-uniform illumination correction algorithm. Both in the spatial domain and frequent domain, the portion of non-uniform illumination and useful information overlap. Therefore, it is important to identify the useful information from the background interference in proposed algorithm.In the paper, several non-uniform illumination correction algorithms are studied. These algorithms can be classified to two types, i.e., spatial domain and frequent domain. In the spatial domain, the calculation is performed to each pixel. While in frequency domain algorithm, signal can be transformed form spatial domain to frequent domain and the further processing is implemented in the frequency domain. Both of them have their own shortcomings. In order to obtain satisfactory result, much calculation is required in spatial domain. While in frequent domain, filters cause rings on the results. Moreover, there is no objective standard to classify the performance of such algorithms.In this paper, a new non-uniform illumination correction algorithm based on linear iteration is proposed to deal with such shortcomings. First, the signal’s deviation from the mean formula is calculated with iteration and characteristic point sets are calculated. For such characteristic points, the average value is used instead to minimize the influence near the edge points in the image. The experimental results show that the proposed algorithm is fast and the useful information is well preserved. The influence of non-uniform illumination is also well controlled.更多还原