Information Technology Reference
In-Depth Information
1
video 2
video 1
clip 1
clip 1
clip 2
0.9
jpeg−q10
960x540
jpeg−q1
960x540
960x540
jpeg−q1
jpeg−q1
0.8
jpeg−q10
jpeg−q10
0.7
man in restaurant
street with trees and bicycle
0.6
0.5
query
0.4
0.3
0.2
0.1
0
1
10
19
28
37
46
55
64
73
GoP label
Fig. 14
GOP retrieval based on spatial descriptors. The query is GOP 1 from clip 1 of version
960
of “Man in Restaurant” (plain curve) or “Street with Trees and Bicycle” (dotted curve).
3.3.3
Temporal Dissimilarity
We now analyze the dissimilarity measure of Eq. (21) using only motion descriptors
(
α
2
= 1). Since the different clips of each sequence in our database differ
from each other mainly with respect to motion information, this measure is expected
to discriminate GOPs of different clips of the same video sequence. This is confirmed
by the experimental results shown in Figure 15, which show the motion dissimilarity
between a fixed query GOP and all GOPs of the two clips of the same sequence as
α
1
= 0
,
0.1
video 1
video 2
0.09
clip 1
clip 1
clip 2
jpeg−q10
960x540
jpeg−q1
960x540
0.08
960x540
jpeg−q1
jpeg−q1
jpeg−q10
jpeg−q10
0.07
0.06
0.05
query
0.04
0.03
0.02
0.01
man in restaurant
street with trees and bicycle
0
1
10
19
28
37
46
55
64
73
GoP label
Fig. 15
GOP retrieval based on motion descriptors. The query is GOP 1 from clip 1 of version
960
of “Man in Restaurant” (plain curve) or “Street with Trees and Bicycle” (dotted curve).
