基于Census变换的立体匹配算法优化及医学影像应用

Research on a fast implementation algorithm based on Census transform stereo matching and its applications in medical imaging

目的:建立并优化一种基于Census非参数变换的立体匹配算法。方法:详细分析了基于Census非参数变换的立体匹配,运用移动窗口技术和高速缓存,采用SSE2指令进行数据流的并行处理从而优化算法结构,缩短取指时间,提高计算效率。用Middlebury标准图像对对该算法进行验证,并用于临床阴道镜子宫颈图像对的立体匹配。结果:快速实现了对Middlebury标准图像对以及子宫颈图像对的立体匹配。结论:该算法能够有效提高立体匹配的实时性,可应用于外科手术导航等计算机辅助外科诊疗系统。

Aim： To propose a fast implementation algorithm based on Census transform stereo matching. Methods：Firstly,the Census transform stereo matching was investigated,and its implementation process was analyzed in detail. Secondly,in order to simplify the calculation process and improve the computational efficiency,the moving window technique and cache were employed to optimize the algorithm structure and shorten the fetch time of the instructions. Finally,the SSE2 instructions were adopted to realize the parallel processing for the data stream,and the proposed algorithm was tested with different image pairs. Results： Fast implementation for the stereo matching with the Middlebury standard image pairs and cervical image pair was achieved. Conclusion： The proposed algorithm can effectively improve the real time of the stereo matching based on Census transform,and it can be applied to the computer-aided surgical diagnosis and treatment system such as surgical navigation.