A_threshold_selection_method_from_gray-level_histograms翻译_中文版

大津二值化算法大津二值化算法（Otsu's method）是一种自适应二值化算法，它是由日本学者大津展之于1979年在日本出版的论文《A Threshold Selection Method from Gray-Level Histograms》中提出的。

算法原理是通过计算图像直方图，尝试找到一个阈值，使得将图像分成前景和背景两部分后，前景和背景之间的方差最小。

找到这个阈值后，将灰度值大于该阈值的像素设为前景，灰度值小于该阈值的像素设为背景。

具体的实现步骤是：1. 统计图像的灰度直方图，即将图像中每个灰度级别的像素的数量记录下来，得到一个长度为256的数组hist。

2. 计算累计直方图函数cumSum，其中cumSum[i]表示灰度值小于等于i的像素的数量之和。

3. 计算每个灰度级别i的像素在前景和背景两个类别中的像素数目分别为n1[i]和n2[i]：n1[i] = cumSum[i]n2[i] = sum(hist) - cumSum[i]4. 计算前景和背景的平均灰度值分别为mean1[i]和mean2[i]：mean1[i] = (cumSum[i]*i)/n1[i]mean2[i] = ((sum(hist)-cumSum[i])*i)/n2[i]5. 计算前景和背景之间的方差（类间方差）：variance[i] = (n1[i]*n2[i])*(mean1[i]-mean2[i])*(mean1[i]-mean2[i])6. 找到使得类间方差最小的阈值，即variance数组中的最大值对应的灰度级别就是最佳阈值。

7. 根据最佳阈值，将灰度值大于该阈值的像素设为前景，灰度值小于该阈值的像素设为背景，得到二值化图像。

优点：自适应性很强，可以适用于各种不同光照下的图像。

缺点：如果图像中的前景和背景类别分布非常不均衡，则可能找不到合适的阈值，导致二值化效果不理想。

最大类间方差法(大津法，OTSU)最大类间方差法是由日本学者大津(Nobuyuki Otsu)于1979年提出的,是一种自适应的阈值确定的方法,又叫大津法,简称OTSU 。

它是按图像的灰度特性,将图像分成背景和目标2部分。

背景和目标之间的类间方差越大,说明构成图像的2部分的差别越大,当部分目标错分为背景或部分背景错分为目标都会导致2部分差别变小。

因此,使类间方差最大的分割意味着错分概率最小。

对于图像I(x,y),前景(即目标)和背景的分割阈值记作T ,属于前景的像素点数占整幅图像的比例记为1ω,其平均灰度1μ;背景像素点数占整幅图像的比例为2ω,其平均灰度为2μ。

图像的总平均灰度记为μ,类间方差记为g 。

假设图像的背景较暗,并且图像的大小为M N ⨯,图像中像素的灰度值小于阈值T 的像素个数记作1N ,像素灰度大于阈值T 的像素个数记作2N ,则有:11N M N ω=⨯ (1.1)22N M N ω=⨯ (1.2)12N N M N +=⨯ (1.3)121ωω+= (1.4)1122μμωμω=⨯+⨯ (1.5) 221122()()g ωμμωμμ=⨯-+⨯-(1.6) 将式(1.5)代入式(1.6),得到等价公式:21212()g ωωμμ=⨯⨯- (1.7) 采用遍历的方法得到使类间方差最大的阈值T ,即为所求。

OpenCV 代码:int myOtsu(const IplImage *frame) //大津法求阈值{#define GrayScale 256 //frame 灰度级int width = frame->width;int height = frame->height;int pixelCount[GrayScale]={0};float pixelPro[GrayScale]={0};int i, j, pixelSum = width * height, threshold = 0;uchar* data = (uchar*)frame->imageData;//统计每个灰度级中像素的个数for(i = 0; i < height; i++){for(j = 0;j < width;j++){pixelCount[(int)data[i * width + j]]++;}}//计算每个灰度级的像素数目占整幅图像的比例for(i = 0; i < GrayScale; i++){pixelPro[i] = (float)pixelCount[i] / pixelSum;}//遍历灰度级[0,255],寻找合适的thresholdfloat w0, w1, u0tmp, u1tmp, u0, u1, deltaTmp, deltaMax = 0;for(i = 0; i < GrayScale; i++){w0 = w1 = u0tmp = u1tmp = u0 = u1 = deltaTmp = 0;for(j = 0; j < GrayScale; j++){if(j <= i) //背景部分{w0 += pixelPro[j];u0tmp += j * pixelPro[j];}else //前景部分{w1 += pixelPro[j];u1tmp += j * pixelPro[j];}}u0 = u0tmp / w0;u1 = u1tmp / w1;deltaTmp = (float)(w0 *w1* pow((u0 - u1), 2)) ;if(deltaTmp > deltaMax){deltaMax = deltaTmp;threshold = i;}}return threshold;}参考文献：Nobuyuki Otsu发表的原文"A Threshold Selection Method from Gray-Level Histograms," Systems, Man and Cybernetics, IEEE Transactions on , vol.9, no.1, pp.62-66, Jan. 1979。

A Threshold Selection Method from Gray-Level Histograms[1][1]Otsu N, A threshold selection method from gray-level histogram. IEEE Transactions on System，Man，and Cybemetics，SMC-8，1978：62-66．一种由灰度直方图选取阈值的方法摘要介绍了一种对于画面分割自动阈值选择的非参数和无监督的方法。

最佳阈值由判别标准选择，即最大化通过灰度级所得到的类的方差。

这个过程很简单，是利用了灰度直方图0阶和第1阶的累积。

这是简单的方法扩展到多阈值的问题。

几种实验结果呈现也支持了方法的有效性。

一.简介选择灰度充分的阈值，从图片的背景中提取对象对于图像处理非常重要。

在这方面已经提出了多种技术。

在理想的情况下，直方图具有分别表示对象和背景的能力，两个峰之间有很深的明显的谷，使得阈值可以选择这个谷底。

然而，对于大多数实际图片，它常常难以精确地检测谷底，特别是在这种情况下，当谷是平的和广泛的，具有噪声充满时，或者当两个峰是在高度极其不等，通常不产生可追踪的谷。

已经出现了，为了克服这些困难，提出的一些技术。

它们是，例如，谷锐化技术[2]，这个技术限制了直方图与（拉普拉斯或梯度）的衍生物大于绝对值的像素，并且描述了绘制差分直方图方法[3]，选择灰度级的阈值与差的最大值。

这些利用在原始图象有关的信息的相邻像素（或边缘），修改直方图以便使其成为阈值是有用的。

另一类方法与参数方法的灰度直方图直接相关。

例如，该直方图在最小二乘意义上与高斯分布的总和近似，应用了统计决策程序 [4]。

然而，这种方法需要相当繁琐，有时不稳定的计算。

此外，在许多情况下，高斯分布与真实模型的近似值较小。

在任何情况下，没有一个阈值的评估标准能够对大多数的迄今所提出的方法进行评价。

这意味着，它可能是派生的最佳阈值方法来建立一个适当的标准，从更全面的角度评估阈值的“好与坏”的正确方法。

二值化可信度模型1. 简介二值化可信度模型是一种用于将连续值转换为二值的模型，在许多应用领域中都有广泛的应用。

例如，在图像处理中，可以使用二值化可信度模型将图像中的灰度值转换为黑白像素，以便进行后续的分析和处理。

在自然语言处理中，可以使用二值化可信度模型将文本中的情感倾向转换为正面或负面情感。

本文将介绍二值化可信度模型的原理、应用场景和常用方法，并提供一些实际案例进行说明。

2. 原理二值化可信度模型的原理是将连续值映射为二值。

在实际应用中，通常是将连续值与一个阈值进行比较，大于等于阈值的被映射为1，小于阈值的被映射为0。

对于不同的应用场景，可以根据需求选择不同的阈值。

例如，在图像处理中，可以根据灰度值的分布情况选择一个合适的阈值，使得图像中的目标物体能够清晰地与背景分离。

在情感分析中，可以根据训练数据的标签分布情况选择一个合适的阈值，使得情感倾向可以准确地被二值化。

3. 应用场景二值化可信度模型在许多领域中都有广泛的应用。

以下是一些常见的应用场景：3.1 图像处理在图像处理中，二值化可信度模型可以将灰度图像转换为黑白图像。

通过选择合适的阈值，可以将图像中的目标物体与背景分离出来，便于后续的目标检测和分析。

3.2 情感分析在情感分析中，二值化可信度模型可以将文本中的情感倾向转换为正面或负面情感。

通过选择合适的阈值，可以将情感分数映射为二值，便于进行情感分类和情感倾向分析。

3.3 信号处理在信号处理中，二值化可信度模型可以将连续信号转换为二进制信号。

通过选择合适的阈值，可以将信号的强度映射为0或1，便于进行数字信号处理和通信系统设计。

4. 常用方法在实际应用中，有多种方法可以实现二值化可信度模型。

以下是一些常用的方法：4.1 固定阈值法固定阈值法是最简单常用的二值化方法之一。

它使用一个固定的阈值将连续值转换为二值。

大于等于阈值的被映射为1，小于阈值的被映射为0。

4.2 自适应阈值法自适应阈值法是一种根据局部像素值的分布情况动态选择阈值的方法。

Guidance System Based on Image ProcessingDahai Yu and Manman ShenSchool of electrical information of Changchun Guanghua University, Changchun, China, 130031 Abstract—To solve the problem of blind travel, a guidesystem based on image processing is designed. The visualinformation collected by a CCD camera. The voice information isused to communicate with the user. The overall design of thissystem is given. Recognize the zebra line by the bipolar systemvalue. Recognize the blind road using saturation histogram andGaussian function.Keywords—blind identification; zebra crossing recognition;image processing; guideI I NTRODUCTIONAccording to the World Health Organization (WHO)statistics in 2010, the total number of people in the world with vision impairment is estimated to be 285 million. 39 million arewhole blind. While China has the most people, the number of blind people is also the most. There is about 5 million blindpeople in China. Due to the physiological defects and the increasingly complex living environment, it brings manyinconveniences to the blind people's life. In view of theinconvenience of blind people, guide dog and guide stick gradually become tools to help blind people travel. However, the guide dog is not easy to train and the cost is high, so the detection range of the blind guide rod is limited.H Wang has designed an interactive guide robot. The robot is composed of haptic device and human-computer interaction system. Analyze the 2D information through the haptic system and transmits the information to the user. But the robot can only detect the surrounding obstacles, not the main effective identification the traffic signs. Zhang Ying designed a guide robot based on embedded technology to identify obstacles and traffic signs. But the robot will not receive the voice message to the blind, which caused a lot of inconvenience to the blind. Han Xuefeng designed an interactive guide robot, through the sensor detection of the external environment, and transfer in the form of voice for the blind. But the robot of traffic sign recognition effect is poor, which can’t meet the actual needs of the blind. Zhang Zhimei et al designed a crawler type guide robot. Use ultrasonic sensor to track the trajectory of black ground preset to avoid obstacles. But it can’t work in no-black environment.In view of the deficiency of the blind guiding technology atpresent, the design of intelligent guide system has greatpractical significance.II S TRUCTURE OF GUIDANCE SYSTEM According to the restriction of the activity of the blind, the function of the guide robot is confirmed, and the overall scheme of the guide system is worked out. As shown in Figure I.FIGURE I.O VERALL SCHEME DESIGNA camera is used to collect the image information of the environment. The data are transferred to DSP. GPS module is used to position the user. Voice module is used to receive or play voices for users. The data are also transferred to DSP. Do some calculate in DSP including traffic signal recognition, Blind identification, zebra crossing recognition and so on.III Z EBRA LINE RECOGNITIONThe zebra belt consists of a group of parallel strips with alternate black and white intervals. The difference of color between black and white stripes is very obvious. Gray contrast is very strong. The rule of black and white alternating is strong. So we can use the bipolar coefficient of the image to characterize and quantify the intensity of the intensity contrast of the zebra line region. Analyze and judge the region with strong gray contrast in the road image by the bipolar system value. If the test area is in the zebra line region, the value of the bipolar system is very high. Otherwise, if the region to be inspected i s not a zebra region, its gray value is basically the same, and the value of the bipolar system is small.First we can use a threshold to segment the image.(a) O RIGIN IMAGE(b) T HRESHOLD SEGMENTATIONFIGURE II.I MAGE SEGMENTATIONThen suppose (μ1,σ1)is the average value of density distribution function of black pixels. (μ2,σ2)is the average2018 International Conference on Advanced Control, Automation and Artificial Intelligence (ACAAI 2018)value of density di stribution function of whith pixels. n1,n2 presents the pixel number of black and white.α=n1n1+n2(1)σ02=ασ12+(1−α)σ22+a(1−α)(μ1−μ2)2(2) The bipolar coefficient γisγ=1σ02[α(1−α)(μ1−μ2)2](3)γ is a value between 0 and 1. When γis 0 , the image is not zebra area. And when γis 1, the image must be zebra area.Then search the edge of the zebra. When there are parallel lines in the image. It could be recognized as zebra lines.(a) T HRESHOLD SEGMENTATION(b) R EGIONS USING BIPOLARITYFIGURE III. R ESULT OF BIPOLAR COEFFICIENT At the end, use the edge of the zebra line. Then we can get the zebra lines info rmation.FIGURE IV. R ESULT OF ZEBRA CROSSING RECOGNITIONIV B LIND ROAD RECOGNITIONThe color of b lind road is usually very bright. So the colors can be used to detect the characteristics of the blind road. This paper gets the blind area using image segmentation. Extract the edge of blind road. First, convert the image from RGB to HSI color space. HSI has three color components. Use the saturation histogram to segment image.(a)O RIGINAL IMAGE(b)H ISTOGRAM OF SATURATION(c) THE SEGMENT IMAGEFIGURE V. R ESULT OF IMAGE SEGMENTATION However, sometimes the color histogram will appear "jagged" shape. It is difficult to divide the image through the peaks and troughs in the histogram. So in order to determine the segmentation point, first use the Gauss smoothing filter to smooth the original color histogram. Suppose S L(x)is the histogram of saturation. The processed color histogram function is:F(x,σ)=S L(x)∗G(x,σ)(4) Here, G(x,σ)is the Gauss function. * presents convolution.This operation not only reduces noise, but also eliminates some tiny saw teeth.FIGURE VI. R ESULT OF BLIND ROAD RECOGNITIONV GPS RECOGNITIONThe remote positioning function can send the blind people to the blind's relatives and friends in real time, strengthen the connection between the blind and relatives and friends, and ensure the safety of the blind. Using SIM808 module and SIM telephone card to realize GPS remote positioning. Through the GPRS flow data of SIM808 GPS upload module programming through the SIM card to the China Mobile networking platform, open platform (OneNET) by computer through the GPS data from the OneNET platform and processed, displayed on the map to achieve GPS remote location positioning module.VI C ONCLUSIONSAiming at the needs of the blind, a guide system is designed. According to the confirmation of the blind activities restrict the function of the system and established the general scheme of the robot; then the realization of recognition, such as the zebra and blind important traffic signs. However, the system needs to be further improved, then to realize traffic lights recognition, intelligent guide system development.R EFERENCES[1]Meng Xiangwei, Yan Xijun, Ouyang, stars, et al. Design of guide barbased on ultrasonic sensors [J]. electronic design engineering, 2012, 20(17): 11-14.[2]Zhang Zhimei, C heng Liying, Zhao Yiheng, et al. Study of guide rob otbased on fuzzy PID control alg orithm [J]. Journal of Shenyang Normal University (NATURAL SCIENCE EDITION), 2015, 33 (1): 81-85. [3]Tang Zhichao, Su Lin, He Chao, et al. Research on traffic sign visualrecognition technology of guide rob ot [J]. compu ter technology and development, 2014 (9): 23-27.[4]Yang Wanhai. Multisensor data fusion and its application [M].Northwest University of Electronic Science and Technology Press, 2004. [5]Xu Yan, Wei Zhen Yu. An improved traffic sign image recognitionalgorithm [J]. laser and Optoelectronics Progress, 2017 (2): 118-125. [6]Khatib O. Real-Time Obstacle Avoidance for Manipulators and MobileRobots[C]// IEEE International C onference on Rob otics and Automation.Proceedings. IEEE, 1986:500-505.[7]Huang Yanbiao, Lu o Guangyue, ho ho. Application of B P neuralnetwork in multisensor data fusion of patrol rob ot [J]. Journal of sensing technology, 2016, 29 (12): 1936-1940.[8]Otsu N. A Threshold Selection Method from Gray-Level Histograms [J].IEEE Transactions on Systems Man & Cybernetics, 2007, 9(1):62-66.。

第3期唐浩等：图像处理方法用于声压敏感核声图的目标定位269Sciences), 2017, 53(1): 135-143.[5]OTSU N. A threshold selection method from gray-level histo-grams[J]. IEEE Transactions on Systems, Man, and Cybernetics, 1979, 9(1): 62-66.[6]申铉京, 刘翔, 陈海鹏. 基于多阈值Otsu准则的阈值分割快速计算[J]. 电子与信息学报, 2017, 39(1): 144-149.SHEN Xuanjing, LIU Xiang, CHEN Haipeng. Fast computation of threshold based on multi-threshold Otsu criterion[J]. Journal of Electronics & Information Technology, 2017, 39(1): 144-149. [7]吴鹏. 萤火虫算法优化最大熵的图像分割方法[J]. 计算机工程与应用, 2014, 50(12): 115-119.WU Peng. Image segmentation method based on firefly algorithm and maximum entropy method[J]. Computer Engineering and Ap-plications, 2014, 50(12): 115-119.[8]董本志, 蔡文宇, 景维鹏. 自然场景图像中叶片区域快速多阈值提取方法[J]. 计算机工程与应用, 2017, 53(22): 198-203.DONG Benzhi, CAI Wenyu, JING Weipeng. Fast multi-threshold extraction method of leaf region in natural scene image[J]. Com-puter Engineering and Applications, 2017, 53(22): 198-203.[9]王祥科, 郑志强. Otsu多阈值快速分割算法及其在彩色图像中的应用[J]. 计算机应用, 2006, 26(z1): 14-15.W ANG Xiangke, ZHENG Zhiqiang. Fast Otsu multi-thresholdsegmentation algorithm and its application in color image[J].Computer Applications, 2006, 26(z1): 14-15.[10]王磊, 段会川. Otsu方法在多阈值图像分割中的应用[J]. 计算机工程与设计, 2008, 29(11): 2844-2845.W ANG Lei, DUAN Huichuan. Application of Otsu' method in multi-threshold image segmentation[J]. Computer Engineering and Design, 2008, 29(11): 2844-2845.[11]刘立, 焦斌亮, 刘钦龙. Otsu多阈值算法推广实现[J]. 测绘科学,2009, 34(6): 240-241.LIU Li, JIAO Binliang, LIU Qinlong. Promotion and realization of Otsu multi-threshold segmentation method[J]. Science of Sur-veying and Mapping, 2009, 34(6): 240-241.[12]刘艳, 赵英良. Otsu多阈值快速求解算法[J]. 计算机应用, 2011,31(12): 3363-3365.LIU Yan, ZHAO Yingliang. Quick approach of multi-threshold Otsu method for image segmentation[J]. Journal of Computer Ap-plications, 2011, 31(12): 3363-3365.[13]朱玲羚. 基于Otsu法的声呐图像多阈值分割方法[J]. 水雷战与舰船防护, 2016, 24(4): 68-71.ZHU Lingling. Multi-threshold segmentation method of sonar image based on Otsu method[J]. Mine Warfare & Ship Self-defence, 2016, 24(4): 68-71.• 简 讯 •中国声学学会水声学分会2019年学术会议在南京召开2019年5月25日，中国声学学会水声学分会2019年学术会议在江苏省南京市召开。

基于图像与加速度传感器的电梯开关门检测张霍【摘要】图像处理技术逐渐应用于各个领域,基于图像处理的电梯开关门检测在电梯安全保护系统和电梯广告投影控制系统等方面被广泛应用.本文提出了一种基于图像和加速度传感器的电梯开关门检测方法,该方法通过图像处理技术判断电梯门开关状态,并结合加速度传感器进行辅助判断.研究结果表明该方法能够实时、准确地检测电梯门的开关状态,具有良好的鲁棒性.【期刊名称】《电子世界》【年(卷),期】2018(000)011【总页数】3页(P16-18)【关键词】电梯;图像处理;边缘检测;加速度传感器【作者】张霍【作者单位】四川长虹电器股份有限公司【正文语种】中文0 引言随着社会经济的不断发展，电梯的使用量不断增加，电梯的开关门故障对人们的出行影响越来越大[1]，实时检测电梯门的开关状态在电梯安全保护系统中至关重要；与此同时，电梯广告投影方式逐渐出现，电梯开关门检测成为其投影控制系统的关键技术。

通过图像传感器采集电梯门区域视频序列，并进行图像处理，能够实时检测电梯门的开关量，为安全保护系统与电梯广告投影控制系统提供依据。

电梯门边界在开关过程中其到中线的距离变化，能够有效表征电梯的开关状态，因此可以通过检测边界特征的位置变化判断电梯门的开关量。

该检测过程可分为3个部分：图像预处理、边界检测和开关判断。

1 预处理经图像传感器采集的原始图像尺寸较大，噪声偏多，直接对其进行处理将影响图像处理的效率和检测结果的准确性，故需对原始图像进行以下预处理。

1.1 检测区域提取在图像采集时，相机位于电梯后侧上方中间位置，与轿厢门相对，并成一定角度俯视。

采集到的图像包含很多无用的背景信息，且下方存在乘客的干扰。

因此，需要将该部分信息去除，以提高图像处理速度和排除干扰。

本文将电梯轿厢门上方40cm高度区域作为待检测区，两门闭合处为检测窗口中心，检测区域宽度为随着电梯门边界变化而调整。

1.2 透视变换摄像头的位置高于轿厢门，视口与轿厢门相对，成一定夹角俯视。