1204 F. Gasparini, R. Schettini / Pattern Recognition 37 (2004) 1201 – 1217
The sRGB values are mapped into the CIELAB color
space [30] (see Appendix A), where the chromatic compo-
nents and lightness are separated [20,25]. The CIELAB is a
perceptually uniform color space in which it is possible to
eectively quantify color dierences as seen by the human
eye; it is widely used in color imaging and is included as a
standard in the international color consortium (ICC) color
proles [28].
(i) We analyze only those pixels with a lightness in an in-
terval that excludes the brightest and the darkest points.
This because the images we consider may already have
been processed during acquisition, and we assume that
imaging device is unknown. Digital cameras often force
the brightest image point to white and the darkest to
black, altering the chroma of very light and very dark
regions. Our experience on a data set of several hundred
images has suggested that we consider the interval of
lightness: 30 ¡L
∗
¡ 95 in identifying color cast. If the
size of the considered portion of image is less than the
20% of the whole, the image statistics are not fully re-
liable. These images are considered unclassiable and
are not processed at all. This is the case of very dark
or very light images, such as those shown in Fig. 2.
Otherwise, the algorithm proceeds with the next step.
(ii) The two-dimensional histogram, F(a; b), of the image
colors in the ab-plane is computed. For a multicolor
image without cast it will present several peaks, dis-
tributed over the whole ab-plane, while for a single
color image, there will be a single peak, or a few peaks
in a limited region (see Fig. 3). The more concentrated
the histogram and the farther from the neutral axis, the
more intense the cast.
The color distribution is modeled using the following
statistical measures, with k = a; b:
k
=
k
kF(a; b)dk; (2)
2
k
=
k
(
k
− k)
2
F(a; b)dk; (3)
respectively, the mean values and the variances of the
histogram projections along the two chromatic axes a,
and b.
(iii) An equivalent circle (EC) with center: C =(
a
;
b
) and
radius: =
2
a
+
2
b
is associated to each histogram.
To characterize the EC quantitatively we introduce a
distance D:
D = − (4)
(where =
(
2
a
+
2
b
)), and the ratio:
D
= D=: (5)
Because D is a measure of how far the whole histogram
(identied by its EC) lies from the neutral axis (a =
0;b= 0), while is a measure of how the histogram
is spread, D
makes it possible to quantify the strength
of the cast.
The algorithm analyzes the color histogram distribution
in the ab chromatic plane, examining its EC and computing
the statistical values D and D
.
1. If the histogram is concentrated and far from the neutral
axis, the colors of the image are thus conned to a
small region in the ab chromatic diagram (Fig. 4). The
images are, instead, likely to have either an evident
cast (to be removed), or a predominant color (to be
preserved), if:
(D¿10 and D
¿ 0:6) or (D
¿ 1:5): (6)
A predominant color could correspond to an intrinsic
cast (widespread areas of vegetation, skin, sky, or sea),
or to a single color close-up (Fig. 5).
To detect images with a predominant color corre-
sponding to an intrinsic cast and a single color close-up,
a simple classier exploiting both color and spatial in-
formation is used [31]. A region identied as probably
corresponding to skin, sky, sea, or vegetation is con-
sidered signicant if it covers over 40% of the whole
image; the image is classied as having an intrinsic
cast, and the cast remover is not applied.
If none of the regions corresponding to skin, sky,
sea, or vegetation occupies over 40% of the whole, but
the image EC is extremely concentrated, D
¿ 6, and
has an high average color saturation, (C
∗
=L
∗
¿ 1—the
ratio between the chroma radius and the lightness is
correlated to the color saturation), the image is clas-
sied as a single color close-up and, also in this case,
the cast remover is not applied.
Images presenting a concentrated histogram which
are not classied as having a predominant color, i.e.
intrinsic cast images or close-ups, are held to have an
evident cast and are processed for color correction as
described in Section 3.2.
2. All images without a clearly concentrated color his-
togram are analyzed with a procedure based on the
criterion that a cast has a greater inuence on a neu-
tral region than on objects with colors of high chroma.
The color distribution of near neutral objects (NNO) is
studied with the same statistical tools described above.
A pixel of the image belongs to the NNO region if
its chroma is less than an initial xed value (set here
at one fourth the maximum chroma radius of that im-
age), and if it is not isolated, but has some neighbors
that present similar chromatic characteristics, than it
belongs to the NNO region. Isolated nearly gray pixels
are probably due to noise. If the percentage of pixels
that satises these requisites is less than a predened
percentage of the whole image, which experience has
suggested to set at 5%, the radius of the neutral region