Automatic Extraction of the Text Drawn on 3-Dimensional Space

本研究选择存在于城市生活空间中的文本，如通知、引导展示等作为机器视觉研究的候选对象，旨在对存在于生活环境中的文本进行提取和获取。因此，假设目标输入图像是在普通照明和摄影条件下拍摄的，在良好的摄影条件下输入的图像不一定得到保证。在此前提下，本研究建立了一种生成适合字符识别的二值级字符图像正面图像的技术，通过自适应计算二值得到的高清二值图像提取一个字符域，利用得到的字符上的排列信息，假设并校正相机拍摄时与目标字符序列之间的三维关系。第一年的研究程度是将人物捕捉在一个环境三维平面下，通过各种灯光、色调和对比度的基础上，可以控制实验室建立的各种照明条件的摄影环境，并开发出不易受光源条件影响的鲁棒二值图像算法。其次，对室内环境下布置的包含字符序列的二值图像建立字符域提取算法；针对字符序列各字符线性排列的特点，对一行左右的字符序列进行了实验，如制导显示，验证了该算法可以从具有复杂背景的字符序列图像的二值图像中提取出字符序列的域。在上述研究的基础上，建立了基于平面（如白板）上的矩形序列从平面对面捕获的图像来假定三维参数的算法。下一财年开发了由三维参数生成右对字符序列图像的算法，并根据上一年建立的二值图像鲁棒生成和鲁棒字符序列提取算法，以及三维参数推定算法，将其统一为一个系统，并在实际环境中进行了良好的实验。少

This research chooses text, which exists in a city life space, such as a notice and a guidance display, as a candidate for research of a machine vision, and aims at the extraction and the acquisition of text, which exist in a living environment. Therefore, it assumes that the target input picture is captured under ordinary lighting and photography conditions, and the picture input in good photography conditions is not necessarily guaranteed. The technique of generating the front face image of binary level character pictures suitable for character recognition was established by standing on such a premise, extracting a character domain from the high-definition binary image obtained by calculating a binary value in adaptation in this research, and presuming and rectifying 3-dimensional relation between the camera at the time of photography, and an object character sequence using the arrangement information on the obtained character. The degree of research first year captured the character … More arranged on an environmental 3-dimensional plane under the photography environment which can control the various lighting conditions built in the laboratory by the basis of various lightings, a color tone, and contrast, and developed robust binary image algorithm which cannot be easily influenced by light source conditions. Next, character domain extraction algorithm was built for the binary image containing the character sequence arranged at indoor environment. Paying attention to the feature of a character sequence with each character being arranged linearly, it experimented to character sequences of about one line, such as a guidance display, and it was checked that the domain of a character sequence can be extracted out of the binary image of the character sequence picture which has a complicated background by the algorithm of this research. The algorithm which presumes a 3-dimensional parameter using the picture captured from across to the plane for the sequence of the rectangle drawn on planes, such as a white board, in parallel to the above-mentioned research was built. The following fiscal year developed the generation algorithm of the right pair character sequence picture from a 3-dimensional parameter, unified it to one system in accordance with the robust formation of binary image and robust character sequence extraction algorithm which were built in the preceding year, and 3-dimensional parameter presumption algorithm, and obtained the result with good conducting an experiment in real environment. Less