In order to improve the real-time performance of 3-channel polarization images processing, a parallel processing method for 3-channel polarization images registration is designed and implemented on the multi-core DSP hardware platform TMS320C6670 based on Fourier-Mellin transform. According to the different data space and correlation degree in images registration, operations of FMT algorithm are optimized as multiple operation tasks with high cohesion and low coupling, and a parallel computing structure based on data stream is designed, which overcomes the problems of low efficiency and high data coupling in the original FMT algorithm. According to the processing characteristics of multiple cores and multiple coprocessors of DSP, the parallel task allocation methods such as grouping isomerism and intra-group isomerism are designed. While load balance of the system is ensured, the multi-parallel computing tasks are allocated accurately and efficiently. The experimental results show that the average registration time is about 16.2 ms under the condition that the registration accuracy is less than 1 Pixel and the rotation error is less than 0.2 degree, which can meet the real-time application requirements of polarization imaging detection.