查询结果:   付泽伟,金城.基于全卷积神经网络的屏幕区域定位算法[J].计算机应用与软件,2019,36(6):128 - 135.
中文标题
基于全卷积神经网络的屏幕区域定位算法
发表栏目
多媒体技术应用
摘要点击数
622
英文标题
A SCREEN AREA LOCATION ALGORITHM BASED ON FULLY CONVOLUTIONAL NETWORK
作 者
付泽伟 金城 Fu Zewei Jin Cheng
作者单位
复旦大学计算机科学技术学院 上海 201203     
英文单位
School of Computer Science, Fudan University, Shanghai 201203, China     
关键词
全卷积神经网络 边缘检测 屏幕定位 深度学习
Keywords
Fully convolutional network Edge detection Screen location Deep learning
基金项目
国家重点研发计划项目(2016YFC0801003) ;上海市科技人才计划项目(17XD1425000)
作者资料
付泽伟,硕士生,主研领域:多媒体信息处理。金城,教授。 。
文章摘要
很多情况下,人们需要记录屏幕、投影仪中出现的信息,但是在拍摄到屏幕的同时不可避免地会拍摄到屏幕外的背景。为了解决这个问题,提出一种在手机等便携设备上找到拍摄视频中出现的屏幕区域的算法。提取出视频中的每一帧;对每一帧用全卷积神经网络得到屏幕边缘图像和屏幕位置图像;在屏幕边缘图像上用直线检测算法检测直线;对屏幕位置图像进行分析,从检测到的直线中找到四条直线作为屏幕区域的边缘。由于全卷积神经网络的加入,该方法不需要设定复杂的参数,而且便于扩展到名片、文档等检测上。实验结果表明,该方法拥有很强的鲁棒性,较好的识别速度和准确率。
Abstract
In many cases, people need to record the information appearing in the screen and projector, but when they shoot the screen, they inevitably shoot the background outside the screen. In order to solve this problem, we proposed an algorithm to find the screen area in the shooting video on mobile phones and other portable devices. We extracted each frame in the video and used full convolution neural network to get the edge image and position image of the screen for each frame. Then the straight line detection algorithm was used to detect the line on the screen edge image, and we analyzed the position image of the screen. Four lines were found from the detected line as the edge of the screen area. Due to the addition of full convolution neural network, this method does not need to set complex parameters, and is easy to extend to business card, document and other detection. The experimental results show that the method has strong robustness, good recognition speed and accuracy.
下载PDF全文