摘要计算机视觉识别一直是近年来的热点问题，新的思路、新的方法不断涌现，但是想要通过一种方法实现绝对稳定卓越的性能还有很长一段路要走，尤其是在手写字符识别领域，手写带来的多样性和不确定性更为研究增加了难度。65297

本文以瓷砖上的手写标号字母识别为具体工程背景，意在实现工人在上游手工为瓷砖标记质量等级字母，计算机通过图像识别在下游自动分类瓷砖，以期进一步深化瓷砖生产线的自动化程度。在研究过程中将数字图像处理的理论知识运用于MATLAB图像工具箱和神经网络工具箱的工程环境下，编程实现了将获取到的流水线上的瓷砖图像预处理和基于灰度阈值的图像切割，得到字母标号部分图像，再提取切割后字母图像的特征向量，送入用样本预先训练好的神经网络感知器模拟，最终实现识别并输出识别结果。本文成功实现了四类等级、深浅两色瓷砖的自动识别，给出了MATLAB的仿真结果图片，在对实验结果进行分析的过程中还就实际工程现场的光照强度、光源位置等问题对识别效果的影响做了进一步讨论，并提出了一些旨在降低误识率的注意事项。

关键词  瓷砖标号 图像分割 神经网络 视觉识别

毕业设计说明书（论文）外文摘要

Title      Study of Real-time Labeled Tiles Recogition        

Abstract

Computer vision and pattern recognition has always been in hot discussion among researchers since its appearance. New thoughts and new methods spring up one after another and each one of them seized the attention and imagination of many scholars. Unfortunately, despite all these breakthroughs metioned above, the pursuit of achieving exellent and steady performance by one single method still has a long way to go and requires our harder-work, especially in the recogniton of handwritten characters. Variety and uncertainty brought by handwritting pose obstacles to the research.

This thesis introduces a labeled tiles recognition system under real engineering background. In the previous process of tiles production line, a capital letter which represents the quality level would be marked by the experienced. And our system in the lower reach intends to classify it automatically using computer vision recognition. Without doubt, this system, once being perfected, can highly increase level of automation in tiles manufacturing process. During the research, image processing toolbox and neural network toolbox of MATLAB, combined with theoretical knowledge of digital image processing, are used to achieve the following functions, pretreatment, label orientation, label segmentation, feature extraction, ANN recognition and eventually classfication result demonstration. After several debugging and testing, the recognition system successfully classify the tiles in both light and dark colors into four different quality levels. All figures of the simulation result are given in the main content. Further discussion about how the illumination intensity and the position of light resource affect the results is recorded as the analysis of the results. Some suggestions and notes are also brought up to improve the system performance.

Keywords  Labeled Tiles      Image Segementation 

          Nerual Network     Vision Recognition