交互式分割算法的可变形目标检测  

Deformable Object Detection Based on Interactive Segmentation Algorithm

作　　者：蒲宝明[1] 霍红[1] 杨东升[1] 李相泽[2] 

机构地区：[1]中国科学院沈阳计算技术研究所,辽宁沈阳110870 [2]东北大学计算机科学与工程学院,辽宁沈阳110819

出　　处：《沈阳大学学报（自然科学版）》2016年第4期296-301,共6页Journal of Shenyang University：Natural Science

基　　金：国家科技重大专项资助项目(2013ZX04007031;2012ZX01029001-002)

摘　　要：针对现有高检测精度算法效率较低的问题,提出一种利用基于水平集的交互式目标分割算法对图像进行预分割的方法,提高了对可变形目标检测的效率.算法首先使用交互式目标分割算法对图像进行初步分割,然后根据分割区域进行部件检测.对于部件模型,使用传统的梯度向量直方图来进行描述.预先对图像进行分割,可以减少检测窗口的检测范围,对于像素较高的图片尤其重要.实验结果发现,与穷举搜索的滑动窗口相比,交互式分割算法在速度方面有明显的提高.As low efficiency in algorithm of detection with high precision,a kind of interactive obj ect segmentation based level-set is proposed to improve efficiency.The proposed algorithm firstly applies interactive obj ect segmentation to preprocessing the image,and then detects every area with part-based model.Traditional histogram of oriented gradients is used to describe a component model.It’s particularly important for high resolution image to segment image in advance for reducing detecting range of the detection window.The results show that,the proposed method has obvious improvement in efficiency comparing with the exhaustive search by sliding window.

关 键 词：目标检测 图像分割 交互式分割 梯度向量直方图 可变形部件模型 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]