Information Technology Reference
In-Depth Information
PAL
MS
1
1
0.95
0.95
0.9
0.9
0.85
0.85
0.8
0.8
0.75
0.75
0.7
Haar
LBP
LRD
LRP
0.7
Haar
LBP
LRD
LRP
0.65
0.65
0.6
0.55
0.6
1
2
3
4
5
6
7
8
9
1
2
3
4
5
6
7
8
9
Fig. 4.
Reduction of detection time (y-axis) when suppressing multiple positions on
single image line by single classifier. x-axis is the number of suppressed positions. Target
error of the suppression classifiers is 5 %.
The tests were performed with four types of image features which have been
shown to perform well in real-time object detection. The features used were
Haar-like features [14] (Haar), Multi-Block Local Binary Patterns [15] (LBP),
Local Rank Differences [5] and Local Rank Patterns [5] (LRP). The real-valued
responses of Haar-like features were normalized by standard deviation of local
intensity and then quantized into 10 bins. The detection classifiers were learned
by WaldBoost [11] algorithm and each contained 1000 weak hypotheses. The
base resolution of the classifiers was 24 pixels wide.
In the first experiment, we focused on what is the the achievable speed-up
using the neighborhood suppression of single and also twelve positions for mod-
erately fast detection classifiers (4.5 - 6 features per position) and moderate
target miss rate (
α
=0
.
05) and also on what is the influence of neighborhood
suppression on precision of the detection. These results are shown in Table 1
and Figure 2. The results indicate large differences between individual image
features. While the average number of weak hypotheses computed per position
was reduced with twelve suppressed positions down to 30 % for LBP and 40 %
for LRP , only 55 % was achieved for LRD and 65 % for Haar-like features. This
can be explained by generally higher descriptive power of LBP and LRP. In gen-
eral, the detection rate degraded only slightly with neighborhood suppression -
by less than 1 % except for all twelve positions and LBP on datasets CMU and
BioID and also LRP on BioID.
We have also evaluated the suppression ability with respect to distance form
the classified position. Figure 3 shows that suppression ability decreases rela-
tively slowly with distance and large neighborhood of radius at least 10 pixels
canbeusedforthetestedLBPandLRPclassifiers.
As mentioned before, single suppression classifier can suppress larger area than
just single position. Relation between speed-up and the size area of suppressed
by a single classifier is shown in Figure 4. The results show that by suppressing
larger area it is possible to reach higher speeds. However, the benefit is lower for
frontal face detection and multiple suppression classifiers would always achieve
higher speed-up.
Search WWH ::
Custom Search