Research on identifying the order of fringe pattern traces using angular scan and zone search method

A method for automatically identifying the order of fringe pattern traces is presented. It uses the simplified Otsu algorithm for obtaining the threshold, the angular scan in the range of 45~ for searching the trace positions, and the zone search technique for identifying different traces. Experimental results show that the proposed method may reliably obtain the order of fringe pattern traces orientating from almost 45° to 90°.     

一种基于荧光信号的海面厚油膜评估方法

油膜厚度是海面溢油污染评估分析的一个重要指标,激光诱导荧光(LIF)技术是目前最有效的海面溢油探测技术之一,基于LIF探测技术的油膜厚度反演算法当下仅有适用于薄油膜(≤10~20 μm)的评估方法,而对于较厚油膜(>20 μm)的评估目前尚无有效的反演算法。鉴于此,提出一种基于LIF技术适用于较厚油膜的反演算法,该算法采用油膜荧光信号反演油膜厚度,推导了油膜厚度反演公式,并给出了基于该反演算法的油膜厚度评估方法。首先采用最大类间方差算法(Otsu)选取合适的荧光光谱波段,然后根据选取波段内每个波长的光谱数据反演油膜厚度,最后采用反演油膜厚度的平均值作为油膜厚度评估结果。研究了该算法的适用范围,给出了该算法有效评估范围最大值与测量相对误差的关系,并结合消光系数给出了在多种测量误差条件下不同消光系数油品有效评估范围的最大值。通过实验对本文方法进行了验证,选用原油和白油的混合油(1∶50)作为实验油品,以波长为405 nm的激光作为激发光源,采集波长范围为420～750 nm,采集了海水背景荧光和拉曼散射光光谱、实验油品的荧光特征光谱和多种不同厚度的较厚油膜的荧光光谱。采用Otsu算法选取420～476 nm波段评估油膜厚度,在实验油品油膜厚度≤800 μm时,该算法对油膜厚度的评估具有较高的精度,平均误差为10.5%;在油膜厚度>800 μm时,平均误差为28.8%,评估误差较大且随油膜厚度的增加快速变大,该实验结果与利用测量相对误差和消光系数的分析结果一致。实验结果表明,该方法可以实现对海面较厚油膜厚度的有效评估,并可以根据测量相对误差和消光系数判断评估结果的有效性。     

基于Canny算子和Hough变换的夜间车道线检测

针对夜间只有车灯照射路面图像整体较暗、光照不均匀、车道线不易检测的问题,提出了一种夜间车道线识别方法。首先,对预处理后的图像采用Laplacian算子进行边缘增强;然后,结合Otsu算法进行Canny边缘检测,再在边缘图像底部1/3区域中利用Hough变换进行直线拟合;最后,在斜率约束的基础上提出了一种内侧车道线提取算法,实现了车辆所在车道内侧车道线的检测。针对多种夜间车道线图像进行实验,结果表明,该算法准确提取出了内侧车道线。提出的方法能克服图像较暗和光照不均的影响,排除旁侧车道线、护栏等的干扰,有助于夜间车辆各行其道。     

基于一维Otsu的多阈值医学图像分割算法

针对传统医学图像分割算法时间复杂度高、 分割精度低等问题, 提出一种基于一维Otsu的自动多阈值分割算法. 考虑到医学图像信息的复杂性, 引入基于梯度、 灰度、 距离的综合信息直方图替代传统的灰度直方图, 并分别赋予这3个信息相应的权值. 采用kd-树作为框架快速自动确定阈值个数, 进而实现Otsu对医学图像的自动多阈值分割. 与最大熵、 基于粒子群优化的Otsu算法等进行对比实验的结果表明, 该算法的分割性能优于其他算法.     

基于模拟退火算法的递归自动阈值分割方法

针对目标图像灰度对比度差的现象,以及对目标对象检测实时性的要求,并考虑到传统的Otsu分割方法在分割图像质量较差以及目标区域小时准确性差的缺点,提出了一种基于模拟退火算法的递归Otsu分割方法。在图像直方图呈双峰的情况下能够准确地找到分割阈值。在成像模糊、光照度较差的情况下此方法仍然可以获得较高的准确度。该方法在保证了检测质量的同时并没有导致运算时间的大幅度提升,有效地保证了处理的实时性。实际应用表明该方法切实可行。     

光电图像光照不均蜕化的校正及阈值分割方法*

分析了CCD获取的图像光照不均蜕化产生原因.针对光照不均蜕化的印刷电路板图像的具体情况,用同态滤波器在频域空间进行滤波而增强图像,从而改善图像的目标与背景对比度.讨论了最大类间方差法,给出了其优点.为了更好的分割图像,在此基础上,考虑两类间距与各类内聚性对图像分割的不同影响,提出了一种改进的最大类间方差法.最后对光照不均蜕化的PCB光电图像用不同方法进行了实验,结果表明改进方法能更好地分割图像.     

基于大津算法和深度学习的开集声纹识别自适应阈值计算方法

针对开集声纹识别的自适应阈值计算问题,提出一种基于大津算法和深度学习的阈值计算方法.首先,采用三层受限Boltzmann机堆叠而成的深度置信网络从Mel倒谱系数中提取语音深层特征;其次,通过Gauss混合模型计算特征的相似度值,对特征的相似度值使用大津算法计算阈值.实验结果表明,该方法识别效果较理想,与等错误率计算阈值...     

汽车轮胎号识别中的预处理问题

为便于对汽车轮胎号字符图像进行识别,对其进行一系列预处理．着重研究胎号图像预处理中的增强、降噪及二值化问题．提出一种将邻域均值滤波法和中值滤波法相结合的图像降噪算法,采用该算法能保护图像边缘和细节．提出一种基于经验知识的改进Otsu算法,可缩小Otsu算法搜索阈值范围,减少计算类间方差的次数．结果表明,胎号增强图像比原始图像清晰,降噪后字符图像边缘不模糊,胎号字符二值化速度得到提高,二值化效果也很好．     

A measurement method for distinguishing the real contact area of rough surfaces of transparent solids using improved Otsu technique

An experimental method of measuring the real contact area of transparent blocks based on the principle of total internal reflection is presented,intending to support the investigation of friction characteristics,heat conduction,and energy dissipation at the contact interface.A laser sheet illuminates the contact interface,and the transmitted laser sheet is projected onto a screen.Then the contact information is acquired from the screen by a camera.An improved Otsu method is proposed to process the data of experimental images.It can compute the threshold of the overall image and filter out all the pixels one by one.Through analyzing the experimental results,we describe the relationship between the real contact area and the positive pressure during a continuous loading process,at different loading rates,with the polymethyl methacrylate(PMMA)material.A hysteresis phenomenon in the relationship between the real contact area and the positive pressure is found and explained.     

Tank segmentation of infrared images with complex background for the homing anti-tank missile

In an effort to achieve fast and effective tank segmentation of infrared images under complex background for the homing anti-tank missile, the threshold of the maximum between-class variance method (i.e., the Otsu method) is experimentally analyzed, and the working mechanism of the Otsu method is revealed. Subsequently, a fast and effective method for tank segmentation under complex background is proposed based on the Otsu method by constraining the image background pixels and gray levels. Firstly, with the prior information of the tank, derive the equation to calculate the number of pixels of tank according to optical imaging principle, and then use the calculated tank size to constrain the image background pixels. Secondly, employ the golden section to restrict the background gray levels. Finally, use the Otsu method to implement the segmentation of the tank. Experimental results demonstrate that the proposed method can get as an ideal result as the manual segmentation with less running time.