DSP平台的图像特征点快速索引结构

针对DSP(digital signal processor,数字信号处理器)平台上的图像特征点匹配问题,提出了一种高效的基于自聚类二分查找树的快速索引结构,并设计了适合于DSP结构特点的索引存储布局。通过在离线情况下将特征点参考数据集逐级地二分聚类,生成多级索引结构。以顺序数组的方式将树状索引结构存储到连续的内存空间中,便于导出为数据文件存储及进一步加载到DSP内存中使用。实验表明,该索引结构能够快速有效地在DSP平台完成特征点匹配工作。

Fast indexing method for image feature matching on DSP platform

This paper proposed a fast indexing method to match image features efficiently on DSP platform. It designed a pro-per layout scheme to load the indexing structure into DSP memories. It used K-means clustering to split the dataset and generate hierarchical indexing data. The index was saved into sequential array to utilize efficient serialization on disk and reconstruction in DSP memories. Experiments show that this method is fast and robust in matching image features on DSP platform compared with other searching method.