简化SIFT算法及其在商标图像检索中的应用

Simplified SIFT algorithm and application in trademark image retrieval

针对商标图像形状简单、颜色单一的特点,提出了一种基于简化SIFT特征的商标图像检索新方法。采用DoG算子在多尺度空间检测图像的关键点,并利用圆环域结构替代SIFT原来的方形结构,对SIFT特征描述符的生成方式进行改进,使其具有计算简单、抗几何畸变性、抗旋转性等优点;然后在关键点匹配过程中,采用RANSAC算法去除错误匹配,从而提高匹配的稳定性与精确性。实验结果表明,该方法比原SIFT方法具有更快的计算速度和更高的匹配精度,能很好地应用在商标图像检索系统中。

According to the trademark image characteristics： simple shape, single color, this paper proposed a new method of simplified scale invariant feature transform（SIFT） feature algorithm for trademark images retrieval. Firstly, used a diffe-rence-of-Gaussian（DoG）function pyramid to identify potential interest points that were invariant to scale, and provided a new descriptor configuration based on a circular window to make it resistant to geometric distortion, rotation and reduce the dimension. In addition, used random sample consensus（RANSAC） to guarantee stability and improved the efficiency of matching. By comparing the experimental results show that this method has faster computing speed, higher precision and do well in trademark image retrieval.