In order to develop a human-computer interface(HCI) for people with impaired or weakened limbs, a novel limbs-free face-computer interface(FCI) was proposed. According to the proposed FCI framework based on facial electromyography(fEMG), this article focuses on the optimization of the fEMG collection locations and facial movements. In order to optimize the collection locations and select actions, mechanism-based analysis and data-based analysis were used. In the mechanism-based analysis, the relationship between the main facial muscles and movements was analyzed in detail. In data-based analysis, a novel method activation degree matrix(ADM) based on fEMG was proposed to analyze the relevant between channels and facial actions. As a result, 8 channels from 32 and 6 movements were picked out for online experiments. After optimizing the collection locations, the features calculation time was reduced by 71.2%, and the accuracy of intent recognition was improved by 5-8%. Finally, some online experiments of FCI-based computer typing application, robotic arm and wheelchair control were successfully carried out.