基于视觉的移动设备屏幕提取与规范化方法  

A Visual Screen Extraction and Normalization Method for Mobile Devices

作　　者：颜硕彦 钱巨[1] YAN Shuo-yan;QIAN Ju(College of Computer Science and Technology,Nanjing University of Aeronautics and Astronautics,Nanjing,Jiangsu 211106,China)

机构地区：[1]南京航空航天大学计算机科学与技术学院,江苏南京211106

出　　处：《计算技术与自动化》2021年第1期124-128,共5页Computing Technology and Automation

摘　　要：为了解决照片中移动设备屏幕图像的获取问题,提出了一种基于视觉的移动设备屏幕图像提取与规范化方法。首先使用轮廓识别算法对原始图像进行边缘检测,获得包含设备和手部的轮廓;其次,通过种子填充法从轮廓得到设备和手部的填充区域;接着在原始图像上使用基于颜色空间的皮肤检测方法得到手部区域,再与之前得到的填充区域做减运算,以得到独立的设备区域;然后再在独立设备区域的轮廓上做多边形拟合,得到设备的四个顶点;最后通过透视变化得到规范化后的设备屏幕图像。实验表明,该方法在背景不复杂且设备作为主体的照片环境下具有较高的准确率。A visual screen extraction and normalization method for mobile devices is proposed to get the mobile devices’screen images from photos.First,we do Canny edge detection to detect the edges of the mobile device.The detected edges may include some parts of the hand holding the device.Then,a seed filling operation is conducted on the image of edges to get the filling area of the whole device and parts of the hand.Next,we do color space based skin detection on the original image to get the filling area of the hand.After that,the area of the device can be calculated by subtracting the filling area of the hand from that of the device and parts of the hand.Four corner points of the device can be got by conducting a polygon fitting operation on the calculated area.Finally,we get the normalized device screen image by perspective transformation according to the corner points.The experiment shows that the method accurately works when the subject device is the main object in a photo and the background of the photo is not complicated.

关 键 词：移动设备 图像提取 边缘检测 皮肤检测 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]