Reducing computational complexity for face detection

Conventional face detection techniques involve computing the classifiers in every sub-window and this demands high computational power. Considering an indoor scenario for vision-based patient monitoring, introducing context-awareness can aid in eliminating redundant computations. A block-based technique aimed at shortlisting sub-windows that give a higher probability of the presence of face by detecting the head-shoulder curves was proposed. The resulting computations in applying the face detection technique to the entire image was compared with the case of performing face detection only within the shortlisted sub-windows. The computational cost analysis shows the significant reduction in computations.