Video Quality Analysis for Concert Video Mashup Generation - Advanced Concepts for Intelligent Vision Systems - page 5

Z = s/m :if m

=0
0: t r is .

(2)

Fig. 2 shows three frames with different amount of blurriness.

(a) Z =3 (b) Z =3 . 9 (c) Z =12

Fig. 2. Example of test frames with (a) low, (b) medium and (c) high blur. The
measured blurriness is given by Z .

2.3 Brightness
In video terms, brightness is related to the visual perception of the amount of
light coming from the display. Therefore, this parameter depends on the lumi-
nance of the scene setting, the contrast of the involved in video processing and
the setting of the display. In a typical indoor concert, the bright areas in a scene
are mainly focused towards the stage and the rest of the scene is poorly lit. In
our observation of concert recordings, another influencing factor is the amount
of burned pixels . Burned pixels represent the pixel values clipped by the maxi-
mum and minimum luminance values 255 and 0, respectively, caused by a very
bright light source against a camera or very dark scenes. Frames containing a
higher amount of burned pictures are generally undesirable as they provide little
color or texture information and produce a very disturbing effect. Frames with a
higher luminance (within the range) and sucient contrast values are associated
with good visibility, pleasant to watch, and colorful images. Therefore, we define
brightness here not in the usual video terms but as a function of the three factors
mentioned above.
If Y is the luminance component of a frame represented in YCbCr colorspace,
the average luminance I l , contrast I c and amount of burned pixels I p of a frame
are given by:

W H

1

I l = Y ( i, j ) , (3)

W × H

i =1 j =1

W H

1

I c = ( Y ( i, j ) − I l ) 2 , (4)

W × H

i =1 j =1

⎛ ⎞

W H

1

⎝ 0 . 1 , ⎠ ,

I p =max p ( i, j ) (5)

W × H

i =1 j =1