Aiming at the real-time requirement of the fabric defect detection systems in industrial application, this paper analyzes the fabric defects detection algorithm via small scale over-complete basis set, and transforms it from VS to TMS320C6678. Moreover, the paper proposes a parallel scheme based on features’ split to make full use of each core’s resources. Incorporating with optimized synchronization and communication method, it achieves efficient multicore parallel processing and reduces the time-consuming greatly. Experimental results show that the method is feasible and effective, and the real-time processing capability of fabric defect detection system based on TMS320C6678 has improved so much, making a solid foundation for future industrial applications.