Low Resolution OCR for Portable Devices to Assist the Visually Impaired

描述（由申请人提供）：CVISION Technologies Inc.提出开发新型光学字符识别（OCR）算法，以提高视障人士的独立性并改善他们的日常生活。该项目将产生OCR软件，允许使用包括手机相机在内的数码相机读取文件、街道标志和其他文本。然后，使用文本到语音的技术，这些文本可以被读回给用户，或者发送到计算机显示或打印在一个易读的大小。在任何一种模式下，该软件都将赋予视障用户更大的独立性，例如，在旅行或签署法律表格时，不需要专门的设备或其他人的帮助。该研究将专门解决低分辨率媒体（如数字视频或数码相机图像）中文本的准确OCR问题。这些设备图像中的低分辨率文本给当前的OCR技术带来了很大的困难。这项工作的核心创新是开发了一种自下而上的OCR方法，而无需首先将图像阈值设置为黑白（即文本和背景）。这与传统的OCR系统相反，传统的OCR系统依赖于存在一个阈值来将文本与背景分开，然后应用自顶向下的方法。自上而下的方法将图像中的文本分解成越来越小的单元（段落、行、词等），直至字形（字符）级别。当采样分辨率是拓扑保持的，通常高于约200dpi时，这工作得很好。建议的自底向上方法以相反的方式工作，首先确定文本的最小单元，然后逐渐增加到越来越大的单元。低分辨率捕获文本中典型的模糊、采样不足的字母之间的模糊边界可以通过放松对灰度图像阈值的需求，然后采用自下而上的方法来有效地处理，而它们会导致传统的OCR算法失败。这种方法的另一个好处是它不需要校准扫描系统（例如均匀照明，水平文本）来成功操作。这将进一步提高CVISION系统对手机和其他非校准源捕获的图像中的文本的准确性。公共卫生相关性：拟议项目与公共卫生的相关性在于其价值，使视障人士能够更独立地生活。例如，想象一下，可以用普通的手机相机拍摄一份视障人士无法辨认的公寓租赁合同，使用本项目开发的软件对合同图像进行OCR，然后让手机通过手机扬声器读取合同。

DESCRIPTION (provided by applicant): CVISION Technologies Inc. proposes the development of novel optical character recognition (OCR) algorithms for increasing the independence of visually impaired people and improving their day-to-day life. The project will result in OCR software that will allow for using digital cameras, including cell phone cameras, to read documents, street signs, and other text. The text can then be read back to the user using text-to-speech technology, or sent to a computer for display or printing at a legible size. In either mode, the software will grant the visually impaired user greater independence while, for example, traveling or signing legal forms, without the need for specialized equipment or the assistance of other people. The research will specifically address the problem of accurate OCR for text in low resolution media such as digital video or digital camera images. The low resolution text in images from these devices gives current OCR technology a great deal of difficulty. The central innovation in this work is the development of a bottom-up approach to OCR without first thresholding the image into black & white (i.e. text and background). This is in contrast to conventional OCR systems, which depend upon the existence of a threshold to separate the text from the background, and then apply a top-down approach. The top-down approach breaks down the text in the image into increasingly smaller units (paragraph, line, word etc.), down to the glyph (character) level. This works well when the sampling resolution is topology-preserving, usually higher than about 200 dpi. The proposed bottom-up approach works in the opposite fashion, identifying the smallest units of the text first and building up to larger and larger units. The vague borders between blurred, undersampled letters that are typical of low resolution captured text can be effectively handled by relaxing the need for thresholding the grayscale image followed by the bottom-up approach, whereas they cause conventional OCR algorithms to fail. Another benefit of this method is that it does not require a calibrated scanning system (e.g. uniform lighting, horizontal text) to operate successfully. This will further increase the accuracy of the CVISION system for text in images captured by cell phone and other non-calibrated sources. PUBLIC HEALTH RELEVANCE: The proposed project's relevance to public health is its value to empower the visually impaired to live their lives with a greater deal of independence. For example, imagine being able to take a photograph of an apartment leasing contract that is illegible to a visually impaired person with an ordinary cell phone camera, using the software developed in this project to perform OCR on the image of the contract, and then having the cell phone read the contract with the phone's speaker.