Digital Signal Processing Reference
In-Depth Information
Fig. 4.10
Some representative results of visual saliency models. (
a
) Original scenes; (
b
) Ground-
truth saliency maps; (
c
)[
24
]; (
d
)[
19
]; (
e
)[
20
]; (
f
)[
68
]; (
g
)[
15
]; (
h
)[
16
]; (
i
)[
13
]; (
j
)[
30
]; (
k
)[
44
];
(
l
)[
46
]; (
m
)MTRL
4.4
Content Analysis for Genre-Specific Video
In this section, we will present two representative content analysis works in two
types of video, i.e., sports video and surveillance video.
4.4.1
Sports Video Scene Analysis
Various innovative and original works have been applied and proposed in the field of
sports video analysis. However, individual works focused on sophisticated method-
ologies with particular sport types and there was a lack of scalable and holistic
framework in this field. This section presents a solution for this issue and presents
a systematic and generic approach which is experimented on relatively large-scale
sports consortia.
4.4.1.1
A Generic Framework for Sports Video Analysis
A system overview from a holistic aspect is illustrated in Fig.
4.11
, such that the
input sports video is analyzed systematically using a generic and sequential frame-
work. This is interpreted such that the result from a preceding process is input to the
next process with a consistent and coherent fashion. The highlights of this frame-
work include:
1. A generic foundation using domain-knowledge free local feature is developed to
represent input sports videos. This method would fit the general framework in
sports video analysis and provides an alternative solution to alleviate generality,
scalability, and extension issues.
