中文摘要：

研究图像区域目标准确分割问题，由于海量图像的特征复杂，传统目标分割算法无法准确提取物体图像的兴趣区域，造成目标的分割准确度较差。为了提高分割精度，提出了一种结合兴趣区域与机器学习SVM的目标自动提取算法，将目标提取作为分类问题，在像素精度范围内对图像进行分割。兴趣区域的提取基于关注窗口思想，首先对图像分层滤波，利用改进的Sojka算法提取目标角点，根据角点位置确定兴趣区域。然后在兴趣区域与背景区域随机提取样本点，根据样本相似性选择用于SVM的训练样本。实验结果表明，基于像素精度的分类方法提高了目标的分割精度，而且目标提取的过程不需要要人工交互，完全实现了目标的自动提取，是处理大量图像目标分割问题的有效方法，为网络图像库的目标自动分割提供了参考。

英文摘要：

The features of large image database is so complicated that traditional algorithms of object segmentation can not extract the ROI（ region of interest） correctly, which causes the inaccuracy of object segmentation. In order to improve the accuracy of segmentation, we proposed a new algorithm of automatic object extraction based on ROI and machine learning algorithm Support Vector Machine （SVM）, which takes the object extraction as a classifying process. The segmentation was conducted in pixel precision, which improved the accuracy of segmentation result. Based on attention window, we used the improved Sojka algorithm to calculate the corner of object which determined the ROI. Then, the samples of object and background were extracted randomly. To improve the classification accura- cy, we used the similarity measure to get rid of the redundancy samples. Experiment results show that the proposed algorithm has higher accuracy in image segmentation. With no human interaction, our method achieved automatic ob- jection extraction, which solves the large image database segmentation problem and provides a reference for object au- tomatic segmentation of image dataset in internet.