【作者】 陈磊；

【导师】 刘芳；

【作者基本信息】 华中科技大学 ， 计算机应用技术， 2015， 硕士

【摘要】 在医药卫生领域,患者在就医的过程中会产生大量的纸质单据。这些纸质单据都将作为重要凭证日后具有多种用途。而目前的纯粹依靠手工处理这些单据存在很多的问题:耗时耗力、效率极低、不便于日后统计与检索等等。在这种情况下,迫切需要一种能够将医疗单据数据快速自动录入的系统,而单据图像中的文字块的分割是其中的一个关键环节。为实现医疗单据图像中文字块的分割,需要完成印章去除、倾斜校正、文字区域提取以及文字块分割等功能。单据中普遍存在印章,这些印章的存在将会给文字块分割带来困扰,采用基于R通道阈值的去除算法,能够有效的将印章去除;单据的扫描图像普遍存在小角度的倾斜,而文本图像检测倾斜角时普遍采用的Hough线变换存在计算量偏大的问题,采用图像子区域选取的自适应算法,能够快速准确定位原始图像中文字最密集的子区域用于倾斜角检测,有效地减少计算量;单据图像文字区域外围普遍存在一些杂点与污渍,采用基于文字轮廓外接矩形与文字尺寸特征相结合的文字区域边界确定算法,能够快速准确地定位单据中的文字区域。最终设计并开发出了一款医疗单据图像中文字块分割系统,并对该系统进行了测试,分别测试了系统的各个主要功能和系统性能。其中,印章去除的正确率达到了88.89%,倾斜校正的正确率达到了95.21%,文字区域裁剪的正确率达到了85.78%,文字块分割的正确率达到了75.86%;在较低的硬件配置下,单张图像处理的平均耗时约为9.399秒。从测试结果来看,基本上可以达到用户的要求。更多还原

【Abstract】 In the medical and health field, the patient medical treatment process will generate a lot of paper documents. These paper documents are very important voucher in future, which can be used in many places. But these documents were handled by manually, which would bring a lot of problems: time-consuming, inefficient, not easy to retrieve statistics in the future and so on. In this case, a fast automatically data input system for medical document is needed urgently, in which, the text blocks segmentation in medical document image is an important step.To achieve text blocks segmentation in medical document image, the seal removing, skew correction, text area trimming, text blocks segmentation and other functions should be implemented. Medical Documents always contain seals, which is harmful for the text block segmentation, the removal algorithm based on R-channel threshold can effectively remove the seal; scanned images of documents always skew small angles, and the Hough Transform which was widely used to detect the skew angle for text images calculate too much, an adaptive algorithm to select image sub-region, can quickly and accurately locate the sub-region in original image for the skew angle detection, and can effectively reduce the amount of calculation; Document images always contain noise and dirt, using the boundary rectangle of the text contour combined with the size of the text algorithm, we can figure out the boundary of the text area, and accurately cut out the text area from the document.The text-block segmentation system in medical document image was designed and implemented, and the system was tested for each of the major functions and system performance. The seal removal rate reached 88.89%, skew correction correct rate reached 95.21%, the text area to trim the correct rate reached 85.78%, and the text block division correct rate reached 75.86%; the average time a single image processing is approximately 9.399 s. From the test results, the system can basically meet the needs of users.更多还原