煤气表数字图像识别算法研究
|
摘要
本文将图像识别技术添加到传统的煤气表检测中,通过对煤气表拍照后的图像进行数字识别从而读取煤气表转动的数据。数字图像识别系统包括图像定位、字符分割、及字符的识别三部分。
图像的预处理主要涉及到图像灰度变化算法,图像的二值化算法,图像的滤波处理算法,图像的锐化处理算法。通过这一系列的变换,得到具有明显特征属性的二值化图像。
图像的定位和分割主要涉及到倾斜校正算法,图像的边缘检测算法。通过对二值化后的图像进行行扫描和列扫描,确定字符纹理区域上、下边界线的水平倾斜角度,据此对图像进行倾斜校正,然后通过投影法确定字符左右边界,进行字符切分,并进行归一化处理。
图像数值的识别主要涉及到图像特征提取算法,特征库的建立算法,特征库比对算法,特征识别逼近算法。数据网络分为个位、十位和百位数字识别的3个子网络。将归一化处理的图像分别送入网络中进行识别,最后把识别出的数值进行组合后得到煤气表显示的数值。
本文将1500个样本作为特征库的收集和识别,识别的成功率可以达到91.67%,算法兼顾了识别率、识别速度、可继承等方面,使得在满足了工程应用的同时也能够方便地被其他应用继承。
This article is about the image recognition technology which is added to the traditional detection of gas meters. The digital is made out by identifying the pictures of the gas meter. Digital image recognition systems include three parts. They are image location, character segmentation, and character recognition.
Image pre-processing mainly relates to the changes of image gray-scale algorithm, two value change of the image, image processing algorithms for filtering, image sharpening processing algorithms. Through these series steps of transformations, we get the binary image with obvious features.
Image positioning and segmentation mainly relates to the skew correction algorithm, the image edge detection algorithm. Through the row scan and column scan of the binarization image, we can determine the bottom level of the boundary angle and correct the image incline. Through the projection method we can determine the borders around the characters. Then character segmentation can be done to normalize the image.
Image recognition mainly relates to the feature extraction algorithm, the character warehouse building algorithm, the character warehouse comparing and classification algorithms, the character value approaching algorithm. The data network is divided into three sub-networks like unit, tens, and centesimal. The normalized images were sent to the network. The results are combined to identify the values obtained in the gas meter image.
In this paper, nearly 1500 samples are sent to the system to learn. The recognition of the success rate can reach 91.67%. The algorithms take into account the recognition rate, recognition speed, inheritable. The application can meet the project and it can easily be inherited by other applications.
图像的预处理主要涉及到图像灰度变化算法,图像的二值化算法,图像的滤波处理算法,图像的锐化处理算法。通过这一系列的变换,得到具有明显特征属性的二值化图像。
图像的定位和分割主要涉及到倾斜校正算法,图像的边缘检测算法。通过对二值化后的图像进行行扫描和列扫描,确定字符纹理区域上、下边界线的水平倾斜角度,据此对图像进行倾斜校正,然后通过投影法确定字符左右边界,进行字符切分,并进行归一化处理。
图像数值的识别主要涉及到图像特征提取算法,特征库的建立算法,特征库比对算法,特征识别逼近算法。数据网络分为个位、十位和百位数字识别的3个子网络。将归一化处理的图像分别送入网络中进行识别,最后把识别出的数值进行组合后得到煤气表显示的数值。
本文将1500个样本作为特征库的收集和识别,识别的成功率可以达到91.67%,算法兼顾了识别率、识别速度、可继承等方面,使得在满足了工程应用的同时也能够方便地被其他应用继承。
This article is about the image recognition technology which is added to the traditional detection of gas meters. The digital is made out by identifying the pictures of the gas meter. Digital image recognition systems include three parts. They are image location, character segmentation, and character recognition.
Image pre-processing mainly relates to the changes of image gray-scale algorithm, two value change of the image, image processing algorithms for filtering, image sharpening processing algorithms. Through these series steps of transformations, we get the binary image with obvious features.
Image positioning and segmentation mainly relates to the skew correction algorithm, the image edge detection algorithm. Through the row scan and column scan of the binarization image, we can determine the bottom level of the boundary angle and correct the image incline. Through the projection method we can determine the borders around the characters. Then character segmentation can be done to normalize the image.
Image recognition mainly relates to the feature extraction algorithm, the character warehouse building algorithm, the character warehouse comparing and classification algorithms, the character value approaching algorithm. The data network is divided into three sub-networks like unit, tens, and centesimal. The normalized images were sent to the network. The results are combined to identify the values obtained in the gas meter image.
In this paper, nearly 1500 samples are sent to the system to learn. The recognition of the success rate can reach 91.67%. The algorithms take into account the recognition rate, recognition speed, inheritable. The application can meet the project and it can easily be inherited by other applications.
引文
[1] Potamianos, Graf, Cosatto. An image transform approach for HMM based automatic lipreading[J]. Issue 5.ieee Volume 4. 1998. Page 4-7
[2] Kaup, Aaeh. Coding of segmented images using shape-independent basis functions[J]. Issue 7. ieee Volume 7. 1998. Page 5
[3] Sangshin Kwak, Yeongwoo Choi, Kyusik Chung. Video caption image enhancement for all efficient character recognition[J]. Issue 6. Ieee Volume 2. 2000. Page 3-7
[4] Xiaoyu Zhao, Chi Xu, Zheru Chi, Gang Chen. Image Recovery from Broken Image Streams[J]. Issue 9. Ieee Volume 3. 2007. Page 2-4
[5] Brejl, Sonka. Object localization and border detection criteria design in edge-based image segmentation: automated learning from examples[J]. Issue 10. Ieee Volume 19. 2000. Page 5
[6] Cors, Hager. Coherent regions for concise and stable image description[J]. Issue 7. Ieee Volume 2. 2005. Page 8
[7] Jian-Jiun Ding, Soo-Chang Pei. 2-D affine generalized fractional Fourier transform[J]. Issue 6. Ieee Volume 6. 1999. Page 4-5
[8] Falkowski, Shixing Yan. Ternary Walsh Transform and Its Operations for Completely and Incompletely Specified Boolean Functions[J]. Issue 8. Ieee Volume 54.2007. Page 3-5
[9]姚敏.数字图像处理[M].修订版2版.北京:机械工业出版社. 2006. 74页
[10]彭玉华.小波变换与工程应用[M].北京:科学出版社. 2005. 33页
[11]朱秀昌,刘峰,胡栋.数字图像处理与图像通信[M].北京:邮电大学出版社. 2002. 54页
[12] Hagan, Martin T. Neural network design[M]. China Machine Press. 2002. Page 49
[13]陈彬彬.高精度手写体数字的识别[C].中国优秀博硕士学位论文全文数据库(硕士). 2006. 17页
[14]胡臣.手写数字的识别及系统的研究[C].中国优秀博硕士学位论文全文数据库(硕士). 1997. 23页
[15]陶茂垣.数字水印中的几何不变特征点提取算法研究[C].中国优秀博硕士学位论文全文数据库(硕士). 2007. 21页
[16]王萍,刘恒,狄光敏.基于简约码特性树的字母和数字识别[J].天津大学学报. 2008. 41页
[17]王亚坤,曾德良,李向菊.一种新颖的数字识别算法[J].北京:电力科学与工程. 2009. 25页
[18]邓士弟,王庆.汽车牌照上英文和数字的识别[R].第五届全国“信号与信息处理”联合学术会议暨陕西省生物医学工程学会二〇〇六年学术年会. 2006. 13页
[19]赵雪春,戚飞虎.基于彩色分割的车牌自动识别技术[D].上海交通大学学报.1998. 32页
[20]魏武,黄心汉.基于模板匹配和神经网络的车牌字符识别方法[J].模式识别与人工智能. 2001. 14页
[21] Gonzalez, Rafael C. Digital image processing[J].Pearson/Prentice Hall. 2007. Page 24
[22]左力.数字图像处理基础知识[DB/OL]. CSDN. 2008. 3页
[23]龚声蓉,刘纯平,王强.数字图像处理与分析[M].北京:清华大学出版社.2006. 36页
[24]吴传孙,邹扬德,周定康.二值化算法在车牌识别中的应用[J].北京:计算机与现代化.2003. 13-16页
[25]谢红,刘玲,刘艳艳.复杂海天线区域检测算法研究[D].北京:应用科技学报.2006. 96-98页
[26]龚昌来.基于数据融合技术的图像均值滤波算法[J].北京:微计算机信息.2007. 313-314页
[27]姜珊,王跃存.基于中值滤波和形态学的去噪算法[J].北京:仪表仪器用户.2007. 106-107页
[28]董付国,原达,王金鹏.中值滤波快速算法的进一步思考[J].北京:计算机工程与应用.2007. 48-49页
[29]芮挺,沈春林,张金林.识别中倾斜牌照的快速矫正算法[J].北京:计算机工程.2004. 122-124页
[30]黎楷模.边缘检测算法在定位中的应用分析[D].郑州:工业高等专科学校学报.2004. 1-2页
[31]李刚,曾锐利.基于数学形态学的定位方法[J].北京:仪器仪表学报.2007. 23-27页
[32]黄豪杰,李榕,常鸿森.基于边缘颜色分布的车牌定位新方法[J].北京:激光杂志.2007. 57-59页
[33]杨俊,戚飞虎.一种基于形状和纹理特征的车牌定位方法[J].北京:计算机工程.2006. 170-171页
[34]陈建坤,范春年.一种基于神经网络的车牌定位方法[D].辽宁:工程技术大学学报.2005. 97-100页
[35]沈庭芝,王卫江,闫雪梅.数字图像处理及模式识别[M].北京:理工大学出版社.2007. 54页
[36]王润生.图像理解[M].北京:国防科技大学出版社.1998. 145-146页
[37] C.von der Malsburg, W.Schneider. A neural cocktail party processor[M].Biol: Cybern. 1984. Page 43
[38] J.L.Johnson, D.Ritter. Observation of periodic waves in a pulse-coupled neural network[J].Opt: Lett. 1993. Page 55
[39] R.A.Kuncheva. Pattern Recognition Method Using two-dimensional Hadamard Transform[M]. Switzerland: Rovisec-7-Advanced Sensor Technology.1988. Page 231-236
[40] S.R. Dubois, F.H. Glanz. An Autoregressive Model Approach to two-dimensional Shape Classification[J]. ieee Trans. PAMI. 1986.Page 56-65
[41] R.L. Kashyap, R. Chellappa. Stochastic Models for Closed Boundary Analysis[J]. ieee Trans. Inform. Theory IT-27. 1981. Page 627-637
[2] Kaup, Aaeh. Coding of segmented images using shape-independent basis functions[J]. Issue 7. ieee Volume 7. 1998. Page 5
[3] Sangshin Kwak, Yeongwoo Choi, Kyusik Chung. Video caption image enhancement for all efficient character recognition[J]. Issue 6. Ieee Volume 2. 2000. Page 3-7
[4] Xiaoyu Zhao, Chi Xu, Zheru Chi, Gang Chen. Image Recovery from Broken Image Streams[J]. Issue 9. Ieee Volume 3. 2007. Page 2-4
[5] Brejl, Sonka. Object localization and border detection criteria design in edge-based image segmentation: automated learning from examples[J]. Issue 10. Ieee Volume 19. 2000. Page 5
[6] Cors, Hager. Coherent regions for concise and stable image description[J]. Issue 7. Ieee Volume 2. 2005. Page 8
[7] Jian-Jiun Ding, Soo-Chang Pei. 2-D affine generalized fractional Fourier transform[J]. Issue 6. Ieee Volume 6. 1999. Page 4-5
[8] Falkowski, Shixing Yan. Ternary Walsh Transform and Its Operations for Completely and Incompletely Specified Boolean Functions[J]. Issue 8. Ieee Volume 54.2007. Page 3-5
[9]姚敏.数字图像处理[M].修订版2版.北京:机械工业出版社. 2006. 74页
[10]彭玉华.小波变换与工程应用[M].北京:科学出版社. 2005. 33页
[11]朱秀昌,刘峰,胡栋.数字图像处理与图像通信[M].北京:邮电大学出版社. 2002. 54页
[12] Hagan, Martin T. Neural network design[M]. China Machine Press. 2002. Page 49
[13]陈彬彬.高精度手写体数字的识别[C].中国优秀博硕士学位论文全文数据库(硕士). 2006. 17页
[14]胡臣.手写数字的识别及系统的研究[C].中国优秀博硕士学位论文全文数据库(硕士). 1997. 23页
[15]陶茂垣.数字水印中的几何不变特征点提取算法研究[C].中国优秀博硕士学位论文全文数据库(硕士). 2007. 21页
[16]王萍,刘恒,狄光敏.基于简约码特性树的字母和数字识别[J].天津大学学报. 2008. 41页
[17]王亚坤,曾德良,李向菊.一种新颖的数字识别算法[J].北京:电力科学与工程. 2009. 25页
[18]邓士弟,王庆.汽车牌照上英文和数字的识别[R].第五届全国“信号与信息处理”联合学术会议暨陕西省生物医学工程学会二〇〇六年学术年会. 2006. 13页
[19]赵雪春,戚飞虎.基于彩色分割的车牌自动识别技术[D].上海交通大学学报.1998. 32页
[20]魏武,黄心汉.基于模板匹配和神经网络的车牌字符识别方法[J].模式识别与人工智能. 2001. 14页
[21] Gonzalez, Rafael C. Digital image processing[J].Pearson/Prentice Hall. 2007. Page 24
[22]左力.数字图像处理基础知识[DB/OL]. CSDN. 2008. 3页
[23]龚声蓉,刘纯平,王强.数字图像处理与分析[M].北京:清华大学出版社.2006. 36页
[24]吴传孙,邹扬德,周定康.二值化算法在车牌识别中的应用[J].北京:计算机与现代化.2003. 13-16页
[25]谢红,刘玲,刘艳艳.复杂海天线区域检测算法研究[D].北京:应用科技学报.2006. 96-98页
[26]龚昌来.基于数据融合技术的图像均值滤波算法[J].北京:微计算机信息.2007. 313-314页
[27]姜珊,王跃存.基于中值滤波和形态学的去噪算法[J].北京:仪表仪器用户.2007. 106-107页
[28]董付国,原达,王金鹏.中值滤波快速算法的进一步思考[J].北京:计算机工程与应用.2007. 48-49页
[29]芮挺,沈春林,张金林.识别中倾斜牌照的快速矫正算法[J].北京:计算机工程.2004. 122-124页
[30]黎楷模.边缘检测算法在定位中的应用分析[D].郑州:工业高等专科学校学报.2004. 1-2页
[31]李刚,曾锐利.基于数学形态学的定位方法[J].北京:仪器仪表学报.2007. 23-27页
[32]黄豪杰,李榕,常鸿森.基于边缘颜色分布的车牌定位新方法[J].北京:激光杂志.2007. 57-59页
[33]杨俊,戚飞虎.一种基于形状和纹理特征的车牌定位方法[J].北京:计算机工程.2006. 170-171页
[34]陈建坤,范春年.一种基于神经网络的车牌定位方法[D].辽宁:工程技术大学学报.2005. 97-100页
[35]沈庭芝,王卫江,闫雪梅.数字图像处理及模式识别[M].北京:理工大学出版社.2007. 54页
[36]王润生.图像理解[M].北京:国防科技大学出版社.1998. 145-146页
[37] C.von der Malsburg, W.Schneider. A neural cocktail party processor[M].Biol: Cybern. 1984. Page 43
[38] J.L.Johnson, D.Ritter. Observation of periodic waves in a pulse-coupled neural network[J].Opt: Lett. 1993. Page 55
[39] R.A.Kuncheva. Pattern Recognition Method Using two-dimensional Hadamard Transform[M]. Switzerland: Rovisec-7-Advanced Sensor Technology.1988. Page 231-236
[40] S.R. Dubois, F.H. Glanz. An Autoregressive Model Approach to two-dimensional Shape Classification[J]. ieee Trans. PAMI. 1986.Page 56-65
[41] R.L. Kashyap, R. Chellappa. Stochastic Models for Closed Boundary Analysis[J]. ieee Trans. Inform. Theory IT-27. 1981. Page 627-637