复杂背景图像的字符识别算法研究

为了解决字符识别过程中的局部曝光、印刷字符的断裂以及变形和自然环境下的背景污染等问题, 提出了一种分块处理与卷积神经网络(CNN)相结合的字符图像识别算法. 首先利用OpenCV机器视觉库, 结合分块处理、伽马运算、参数调整等方法对产品零件表面印刷字符进行预处理, 初步解决图像局部曝光和字符断裂问题; 其次为了获得单个字符图像, 利用数学形态学算法对局部曝光处理后的二值化图像进行分步分割, 进而去掉字符间的无用信息; 最后利用Keras模块为字符识别提供的API搭建CNN模型, 经过对100多张字符的识别训练, 准确率高达96.9%, 为某汽车零部件自动化生产中的字符识别提供了可靠的依据.

Automatic Recognition Algorithms Research for Printed Characters with Local ExposureZhang Hongxia 1, Wang Can 1, Liu Xin 1, Bai Zhicheng 1, Fu Xiujuan 1, Wang Gang2 , Mei Tiancan2 ,

To solve the problems of local exposure, fracture and distortion of printed characters, and background pollution in natural environment during character recognition, a character image recognition algorithm combining block processing and convolution neural network (CNN) is proposed. Firstly, using the OpenCV library, the printed characters on the surface of product parts were preprocessed by block processing, gamma operation and parameter adjustment, and the local exposure and character breaking of images were preliminarily solved. Secondly, in order to obtain a single character image, the image processed in the early stage is multiple segmented by the mathematical morphology algorithm, so as to remove the useless information between characters. Finally, the API provided by Keras for character recognition was used to build CNN model. After over 100 characters recognition training, the accuracy rate was as high as 96.9%, the model provides a reliable basis for character recognition in automatic production of an auto parts.