结合MSER与HSOG的目标局部特征提取

Target local feature extraction combined MSER and HSOG

针对SIFT在视点变化下对物体几何特征描述的缺陷,本文提出一种结合最稳极值区域检测和改进的二阶梯度直方图描述子的目标局部特征提取方法。首先,采用一种新的最稳定度判断准则,提高了不规则形状区域和模糊条件下的检测效果;然后,利用改进的二阶梯度直方图提取MSER区域的局部特征描述子,采用高斯函数加权的方法考虑了不同像素对区域中心像素的影响,提高了稳定性;最后,通过标准测试图像和实际拍摄图像的匹配对算法进行验证。实验结果表明,本文方法在视点变化下仍能获得70%以上的匹配率,匹配效果优于SIFT。本文方法相比于传统方法检测效果更为稳定,对于不规则形状区域仍有较好的检测效果,适用于视点变化下的目标匹配。

For the defect in describing object geometry features of SIFT at viewpoint variation,a new local feature extraction method is proposed in this paper,which combined maximally stable extremal region(MSER)and histogram of second order gradient(HSOG).First,a new most stability criterion is adopted to improve the detection effect at irregular shaped regions and under blur conditions.Then,the local feature descriptors of MSER is extracted by the improved histogram of second-order gradient;the influences of different pixels on central pixel of MESR region are considered by weighted Gaussian function method and the stability is improved.Finally,the method proposed is verified through image match with standard test images and real images.Experimental results show that the method proposed can still achieve more than 70% matching rate at different viewpoint,which is better than SIFT.Compared with the traditional method,the presented method achieves an excellent detection effect for irregularly shaped areas,which is suitable for target matching when viewpoint is current.