Efficient face detection method with eye region judgment

EURASIP Journal on Image and Video Processing

Table 2 Computation complexity comparisons

Operation number	Method
	Traditional Adaboost face detector with Haar-like features		The hybrid Adaboost-based face detector of PFMPF and PBHFA		Proposed system
Addition/subtraction (suppose a detection window of size N × N)	Integral image	3 × N × N	PFMPF	N × N	JEER for PFMPF	Less than (N × N) + 0.1(N × N)
	K weak classifier	7 × K (suppose all the K weak classifiers are two-rectangle features)	PBH feature	Less than 2 × N × N	JEER for PBH feature	Less than (2 × N × N) + 0.1(N × N)
Addition/subtraction (suppose a detection window of size 24 × 24)	Case 1: 1,847 operations (suppose 17 weak classifiers are two-rectangle features)		85% of images use 576 operations, and 15% of images use less than 1,728 operations		91% of images use 633 operations, and 9% of images use less than 1,843 operations
	Case 2: >1,847 operations (suppose three-rectangle or four-rectangle features are used)

Comparisons among the proposed system, the hybrid Adaboost-based face detector of PFMPF and PBHFA, and the traditional Adaboost face detection system that uses the Haar-like features.