图像式水表读数识别方法研究

【作者】 吴梦麟；

【导师】 夏德深；

【作者基本信息】 南京理工大学 ， 计算机软件和理论， 2006， 硕士

【摘要】 随着信息社会的发展，图像处理和机器视觉技术在各行各业的应用越来越广泛。各行业的管理手段正在从人工管理向自动或半自动方向转化。本文提出图像式水表读数的自动识别方法，就是用计算机读数来代替过去抄表员的人工抄表，以达到提高效率和降低人工成本的目的，具有广泛的理论和应用价值。 本文的主要工作包括： (1) 从图像预处理入手，首先针对光照造成的图像灰度不均匀，经典的阈值分割方法如Ostu、Bernsen算法不能很好提取目标的问题，采用了利用梯度点信息插值的S．D．Yanowitz算法，能够很好的分割目标和背景；再利用水平和竖直投影提取水表表框信息；然后再利用改进的Hough变换提取表框范围内直线，对结果进行统计分析，求出水表的倾斜角度；最后采用逆仿射变换的方法，对表框图像进行双线性插值旋转，得到表框的矫正图像。 (2) 针对水表图像的读数特点，提出逐步求精的策略进行数字字符的精确定位。先采用先验知识对字符进行粗定位；再采用形态学的方法对图像进行增强，并对连通域筛选，以提高精确定位的精度；最后利用投影的方法，求出字符的外接矩形的切线位置，并按照半字和整字的不同方法定位，取得良好的定位效果。 (3) 采用基于改进Hausdorff距离的滑动模板匹配进行数字字符识别。利用了分形理论中的Hausdorff距离来计算字符与模版的相似程度，并且采用了改进的Hausdorff距离以提高算法的稳定性。在识别半字的过程中研究了滑动模板匹配的方法，解决了半字模版不唯一的问题。实验证明，算法获得较好的识别率。更多还原

【Abstract】 With the development of information society, image processing and machine vision is widely applied to all walks of life. All trades and professions management tool converts to automation or semi- automation from manual work. In this paper, we propose a method for water meter reading recognition, that is to say, computer reading substituting for manual reading in the past, can improve efficiency and save labor cost, and it has extensive academic and utility value.The work including:(1) This paper introduces arithmetic for image pre-processing. First, because of illumination, the image intensity is not uniform. The classic binarization algorithms such as Ostu algorithm and Bersen algorithm can not extract object well, so we use S.D.Yanowitz algorithm which interpolates the image grey by gradient imformation. Then it can get water meter frame information by horizontal and vertical projection, and extract line by improved Hough transform. Finally, the image can be corrected by affine transform.(2) This paper presents a step wise refinement strategy for digit character location. First, it can use a priori knowledge for approximate location. Second, morphologic enhancement and connected component filter make location more precisely. At last, the tangent rectangle can be solved by projection method, and can be differentiated between complete character and half character.(3) This paper researches on a digit character recognition method that is gliding template matching based on improved Hausdorff distance. Using Hausdorff distance in multi-fractal theory calculates the similarity between image and template, and improves the arithmetic’s stability. In the process of the half digit character recognitions, the paper proposes sliding template concept, solves the problem that it has no uniqueness shape. Experiment proves the recognition rate of the arithmetic meets the practical need.更多还原