用户感兴趣区域图像的自动定位是图像目标自动裁剪和内容检测识别的重要基础和关键之一.论文设计出一套基于二值图像连通域特征的区域定位算法:通过连通域分析,去除图像中与目标无关的内容,提取出候选区域;使用最小二乘法拟合区域进行顶点搜索运算,实现对图像中呈现为凸四边形的目标区域定位;对于定位结果,采用基于消隐点的图像几何校正,有效消除透视畸变.在自主研发的Android平台照片笔记APP中对该算法进行了矩形目标图像自动裁剪测试,结果表明其定位的准确率高于基于边缘特征的定位方法,能得到更准确的裁剪结果,特别是输入图像背景复杂或存在透视畸变情况下,该算法具有明显优势.且该算法执行复杂度低(耗时不到对比方法的1/5),易于优化,更适合在移动设备中应用.%The automatic positioning of the user interest image region is one of the important foundation and key of image tar?get automatic content detection and recognition.A set of regional positioning algorithms based on the connected domain characteris?tics of the binary image is proposed.The content of the image is removed and the candidate regions are extracted by the connected domain analysis;Using the least squares fitting region to carry on the vertex search operation,the target region positioning present?ed as the convex quadrilateral in the image is realized.For the positioning results,the image geometric correction based on vanish?ing point is adopted to effectively eliminate the perspective distortion.In the self-developed photo-notes app of Android platform, the algorithm is used in the target image area automatic cutting test.The results show that the positioning accuracy of the algorithm is higher than the method based on Hough transform,then more accurate cutting results can be getten,especially when the input im?age has complex background or has perspective distortion,this algorithm has obvious advantages.Moreover,the algorithm has low execution complexity(less than 1/5 of the comparison method),which is easy to be optimized and more suitable for application in mobile devices.
展开▼
