Abstract: This paper introduces a matting model based on exemplar to realize automatical training and detection in real-world image. Combined with spatial histogram of visual words improve the performance of order less bag-of-features image representation. Background and foreground information can be got automatically from the phase of localization. It uses graph-cut technique to accomplish matting automatically. The test results on Pascal VOC2006 test set show that the detection rate of this method is high, and its automatically matting effect is as well as existing interactive image cutout tool.

Key words: object detection, matting, bag of feature, object classification

摘要： 引入基于例子的抠图模型，实现对自然图像的自动训练和检测，采用视觉单词的层次空间直方图改进特征包分类检测技术，提高自动检测和定位的精度。从检测结果中自动提取背景信息和前景信息，使用graph-cut技术完成自动抠图。对Pascal VOC2006测试集的测试结果表明，该方法的检测率较高，其自动抠图效果达到了现有交互式图像提取工具的水平。

关键词: 对象检测, 抠图, 特征包, 对象分类

CLC Number: 

• TP391.41