Orientation and scale invariant scene matching with high speed and performance

Abstract

針對場景匹配技術中以二進制穩健基元獨立特征 (binary robust independent elementary features, BRIEF) 為代表的實時算法匹配錯誤率高的問題, 提出一種基于局部梯度二值化的特征描述算法. 該算法利用重心向量方向歸一化特征描述區域, 保證了特征描述符的方向不變性. 同時, 融合基于局部梯度二值化的區域紋理信息以降低特征匹配錯誤率. 使用國際通用數據庫對算法進行了驗證, 實驗結果表明: 提出的場景匹配算法其平均匹配準確率比 BRIEF 算法高 44.59%, 具有較高的魯棒性. For real-time scene feature extraction and matching, conventional binary descriptors improve the speed of the descriptor generation and matching procedure while the false matching rate is high, such as Binary Robust Independent Elementary Features (BRIEF) which is only based on pixel intensity comparisons. To solve this problem, an improved binary descriptor was proposed in this paper, which preserved not only the pixel intensity information, but also the local texture information based on the gradient value. Additionally, the orientation of the centroid vector was also used in the descriptor calculation process, so that the binary descriptors were orientation-invariant. Image Sequences dataset was used to evaluate the performance of the proposed method, and the average matching accuracy rate of the proposed method was 44.59%, higher than that of the BRIEF algorithm. Experimental results show that the proposed descriptors have high accuracy and robustness when dealing with image rotation and scale transformation.