Improving face detection with bi-level classification model

The event of unrest in the three southern provinces has long been a non-trivial burden to national security. The series of harmful acts causes a high-toll lost in both official and civilian lives, yet damages to personal as well as public possesssions. For some time, the government, espcially the Ministry of Defence, pays a great deal of effort to resolve the situation and restore peace in the area. This proves to be effective only to a certain extend, while car bombs and attacks on officials continue. The overall strategy set for this problem still lacks the synergy of advanced technology made available in academic and research communities. Intelligence acqusition and sharing seems to be the weak link in the ideal problem-solving scheme. This article reports an improvement made towards such end, with face recognition being exploited to identify individuals from CCTV images. The precision of this tool depends highly on the detected image area that is thought to be a human face. A benchmark technique to face detection is prone to errors, due to quality of magnitude of sample face positive and negative samples. To overcome this, a new bi-level model is proposed to improve accuracy and reduce the amount of false positives.