摘要
自然场景中的字符识别有很多有意义的应用,比如机器人的自动导航,场景中文字的即时翻译等。深度相机在机器人以及可穿戴设备中已经有较为广泛的应用,而深度信息是否能辅助字符检测还没有被研究。一种基于SWT(Stroke Width Transform)的面向RGB-D图像的字符检测系统在这里被介绍,该算法利用场景三维结构和字符分布特征来优化SWT进行字符检测。虽然利用深度信息限制了该项研究的应用领域,很多应用仍然能够从该项研究中获益:比如携带Kinect的机器人的自动导航,增强现实眼镜Hololens的即时翻译等等。实验证明通过深度信息的辅助能够显著地提高基于SWT字符检测系统的性能。
Automatic detection and recognition of text in the natural scene is a prerequisite for a couple of applications,such as automatic robot navigation and instant scene text translation system.The RGB-D camera has wildly used in mobile robot and wearable device,whether depth can benefit text detection,however,has not been investigated deeply.A text detection system based on SWT using RGB-D image is introduced here,and depth Channel of RGB-D image and distribution of text are used to optimize the performance of SWT.Even though the dependency of Depth information would limit the application,many applications still can benefit from this research,such as automatic navigation of robot equipped with Kinect.and instant translation system of Holo Lens.The experiment result shows that the depth channel indeed can promote the performance of SWT text detection system.
出处
《微型电脑应用》
2015年第9期33-36,5,共4页
Microcomputer Applications
基金
国家自然科学基金面上项目(61175036)
