针对二进制描述子主方向计算误差影响图像匹配正确性的问题,提出了一种旋转不变的二进制描述子构造和匹配方法。在以特征点为中心的同心圆周上等间隔选取采样点,按照旋转不变的模式选取采样点对进行灰度值比较,将比较结果连接成二进制串,作为区域的描述子。匹配时比较描述子在所有旋转模式中的汉明距离,取最小值作为描述子的距离,实现不依赖于主方向的旋转不变性。配合多尺度特征检测方法,将采样模式进行缩放,并对图像进行亚像素插值,实现尺度不变性。描述子匹配的实验结果表明,本文方法在旋转、尺度和光照变换下的匹配效果比当前依赖主方向的二进制描述子方法具有更高的鲁棒性。

Rotation invariance of existing binary local descriptors is achieved by constructing the descriptor relative to a dominant orientation, which is a major source of matching errors. A novel rotation-invariant binary local descriptor is proposed together with its matching method. Sampling points are equally spaced on circles concentric to the keypoint. The intensity comparison results of these sampling point pairs are concatenated into binary descriptor in a rotation-invariant ma- nner. By matching for all rotations, the descriptor is rotation-invariant. Combined with a multi-scale feature detector, scale invariance can also be achieved by sub-pixel sampling. Experimental results on standard evaluation datasets show that the proposed descriptor outperforms other state-of-the-art binary descriptors under rotation, scale or illumination transformation, as the error-prone dominant orientation calculation is avoided.