Low quality text recognition based on machine vision and neural network
                                                     LI Shaohui, ZHOU Jun,LIU Bo, QIAN Yuhao, WU Minyi
                        (School of Mechanical and Electrical Engineering, Hohai University, Changzhou 213000, China)

Abstract: Aiming at the problem of the text picture on the assembly line product contains more noises and defects, which results in the low accuracy of the character recognition and the poor robustness of the machine vision product, the image preprocessing, character segmentation and normalization, character recognition in the text recognition were studied, By using the preprocessing method based on affine transformation in machine vision technology, the text image was skewed to ensure the accurate segmentation of subsequent characters. An improved BP neural network algorithm was proposed to improve the accuracy and robustness of character recognition. By using the additional momentum method and the adaptive learning rate method, the traditional BP neural network was easy to fall into local minima and the convergence speed of neural network model was improved. The results show that the proposed method can effectively correct the skew image and maintain high recognition rate and robustness in the character picture with noise and defect.

Key words: machine vision; assembly line; text recognition; tilt correction; improved BP neural network