Cascade of Boolean detector combinations

EURASIP Journal on Image and Video Processing

Table 3 Performance comparison of different BOA cascades

BOA	F ₁	CT
d₁=(l₁≥θ₁)	.674	0.01
d₂=(l₂≥θ₂)	.525	0.43
d₃=(l₃≥θ₃)	.553	184
B_C2,N_q = 1 ∀q	.764	0.02
B_C2,N_q by BOATS	.778	0.44
B_C3,N_q = 1 ∀q	.774	2.5
B_C3,N_q by BOATS	.813	6.9
¬B_¬OR	.763	4.1
\(\neg \mathrm {B}_{\neg \mathcal {P}},N_{q}\,=\, 1\,\forall q\)	.813	7.0
B_AND,N by BOATS	.814	8.5
B_OR	.683	182.0
\(\mathrm {B}_{\mathcal {P}},N_{q}\,=\, 1\,\forall q\)	.798	153.3
\(\mathrm {B}_{\mathcal {P}}, N_{q}\) by BOATS	.817	124.3

Results in terms of F₁-score and computation time (CT) in respect to video time in scene detection task with detectors d₁, d₂, d₃ and BOA combinations of (17) and (18). The BOA thresholds are selected with the proposed BOATS algorithm. The results are test averages over sixfold cross validation sets. The used operating point of each BOA is the one with highest F₁-score on train data separately for each CV-fold