基于采样点组二值化策略的鲁棒二值描述子研究

发布时间：2018-08-03 17:08

【摘要】：鉴于当前基于采样模型的二值描述子的采样信息相关度高且描述子的鲁棒性较低,通过改进视网膜采样模型,提出基于采样点组二值化策略的鲁棒二值描述子。首先,通过减少采样层数并增大采样点间的距离,设计出低采样点密度和低采样区域重叠度的改进视网膜采样模型。然后,在模型中的采样点圆形邻域内获取若干像素点,将其与采样点一起组成采样点组,分别计算两个采样点组对应点的灰度对比结果,并利用投票策略决定最终二值结果。最后,将采样点组的梯度对比信息与灰度对比信息一起编码生成描述子,以提高对相似灰度区域的描述力。通过对比实验可以看出,所提二值描述子对各种图像变化具有较好的鲁棒性且具有较好的匹配效果。
[Abstract]:In view of the high correlation of sampling information and the low robustness of the descriptor based on the sampling model, a robust binarization descriptor based on the binarization strategy of sampling points is proposed by improving the retinal sampling model. Firstly, by reducing the number of sampling layers and increasing the distance between sampling points, an improved retinal sampling model with low sampling point density and low sampling area overlap is designed. Then, a number of pixel points are obtained in the circular neighborhood of the sample points in the model, which are combined with the sampling points to form the sampling points group. The gray contrast results of the corresponding points of the two sample points groups are calculated respectively, and the final binary results are determined by voting strategy. Finally, the gradient contrast information of the sample group is coded together with the gray contrast information to generate a descriptor to improve the describing power of the similar gray area. The comparison experiments show that the proposed binary-valued descriptor is robust to various image changes and has a better matching effect.
【作者单位】：河南理工大学计算机科学与技术学院;
【基金】：国家自然科学基金(61472119,61572173,61472373,61401150) 计算机视觉与图像处理创新团队(T2014-3) 河南理工大学杰出青年基金项目:基于二值特征描述子的图像匹配方法研究(J2016-3)资助
【分类号】：TP391.41

【相似文献】