Zaharescu__Eugen Color Image Indexing Using Mathematical Morphology

An. S¸t. Univ. Ovidius Constant¸a Vol. 22(1), 2014, 281–288
Color Image Indexing Using Mathematical
Morphology
Eugen Zaharescu
Abstract
A mathematical morphology based approach for color image index-
ing is explored in this paper. Morphological signatures are powerful
descriptions of the image content in the framework of mathematical
morphology. A morphological signature (either a pattern spectrum or
a differential morphological profile) is defined as a series of morphologi-
cal operations (namely openings and closings) considering a predefined
pattern called structuring element. For image indexing it is considered
a morphological feature extraction algorithm which includes more com-
plex morphological operators: i.e. color gradient, homotopic skeleton,
Hit-or-Miss transform. In the end, illustrative application examples of
the presented approach on real acquired images are also provided.
1 Introduction
Morphological image analysis based on Mathematical Morphology (MM) the-
ory uses the lattice theory, the set theory and Euclidian geometry to investigate
the image spatial structures, the shape features of the image objects and the
relationships between them ([6], [7], [8]). Based upon the MM theory, the
precise detection of the object’s pixels along with pertinent indexing features,
proved to be more computationally efficient than other approaches (i.e. sta-
tistical methods concentrate on individual pixel values). But the extension of
mathematical morphology for multivariate functions or multichannel images
(e.g. color images) is a challenging approach. The definitions of the whole
pyramid of the morphological operators starting from the basic ones (erosion
Key Words: Morphological signature, morphological feature extraction algorithm, color
gradient, homotopic skeleton, Hit-or-Miss transform.
Received: November, 2013
Accepted: January, 2014
281
DOI: 10.2478/auom-2014-0024

282 Eugen Zaharescu
and dilation) up to the next upper levels of derived operators (opening, clos-
ing, skeleton, top-hat, etc.) use a totally ordered complete lattice structure.
This algebraic structure cannot be defined in a naturally or perceptually cor-
rect way onto the vector space of color images. The vector mathematical
morphology theory have been analyzed in some fundamental papers ([1], [2],
[3]). Several different orderings have been proposed for the vector space of
color images:
• Marginal ordering uses the artificial point wise ordering (i.e. it orders
color vectors component by component independently). In this case the
disadvantage is that ”false colors” (new color vector values that are not
present in the input image) can be introduced in the processed image.
• Conditional ordering performs the ordering of color vectors by means of
some marginal components selected sequentially according to different
conditions (i.e. lexicographic ordering). The conditional ordering pre-
serves the input color vectors and has been studied especially for HSV
(Hue/Saturation/Value) representation of color images.
• Reduced ordering realizes the ordering of color vectors according to some
scalars, computed from the components of each vector with respect to
different measure criteria, typically distances or projections. The re-
duced ordering has been used to define morphological operators, in the
framework of color morphology, by means of distances. It can be suc-
cessfully used in filtering applications as well as the conditional ordering.
([9],[10]).
For the morphological operators we will use the classical functional defi-
nition from [5] and [6], where the morphological erosion (εg(f)) and morpho-
logical dilation (δg(f)) are respectively defined as follows:
εg(f)(x) = (f⊖g)(x) = inf{f(x−y)−˘g(y) | y ∈ Supp(g)}, ∀x ∈ Supp(f) ⊆ Rn
(1)
δg(f)(x) = (f⊕g)(x) = sup{f(x−y)+˘g(y) | y ∈ Supp(g)}, ∀x ∈ Supp(f) ⊆ Rn
(2)
where the grayscale image f : Ef ⊆ Rn
→ R and the structuring element
or the structuring function g : Eg ⊆ Rn
→ R are semi-continuous functions
and Supp(f) = Ef and Supp(g) = Eg are the definition domains for each
function (in the following experiments Supp(f) and Supp(g) are Euclidian
spaces, Ef , Eg ⊆ R2
or Ef , Eg ⊆ Z2
, in case of an integer grid). Also ˘g is the
symmetric structuring function defined as follows: ∀x ∈ Rn
, ˘g(x) = g(−x).
In case of 2-D binary images these basic morphological operators binary
erosion (⊖) and binary dilation (⊕) are respectively rewritten more simplified:
A ⊖ B = {(x, y)|B(x,y) ⊆ A} =
∩
(u,v)∈B
A(−u,−v) =
∩
(u,v)∈BS
A(u,v) (3)

Color Image Indexing Using Mathematical Morphology 283
A ⊕ B = {(x, y)|Bx,y
∩
A ̸= {ϕ}} =
∪
(u,v)∈B
A(−u,−v) =
∪
(u,v)∈BS
Au,v (4)
where A is a 2 − D binary image and B is a 2 − D structuring element
defined on an Euclidian space, E ⊆ Z2
. A(u,v) and B(x,y) are the translations
of A and B by vectors (u, v) and (x, y). BS
denotes the symmetric structuring
element B:
BS
= {(x, y) ∈ E|(−x, −y) ∈ B}. (5)
Subsequently, a multitude of MM operators are derived from dilation and
erosion such as morphological opening:
γg(f)(x) = δ˘g(εg(f))(x) = ((f ⊖ g) ⊕ g)(x), ∀x ∈ Ef ⊆ Rn
(6)
and morphological closing:
ϕg(f)(x) = ε˘g(δg(f))(x) = ((f ⊕ g) ⊖ g)(x), ∀x ∈ Ef ⊆ Rn
(7)
and morphological gradient:
▽g(f)(x) = δg(f)(x)−εg(f)(x) = ((f ⊕g)(x)−(f ⊖g)(x), ∀x ∈ Ef ⊆ Rn
(8)
where the structuring function g : Eg ⊆ Rn
→ R and the symmetric structur-
ing function ˘g(x) = g(−x), ∀x ∈ Rn
, are the same semi-continuous functions
defined above. In case of 2 − D binary images morphological binary opening
(◦) and binary closing (•) are respectively defined as:
A ◦ B = (A ⊖ B) ⊕ BS
= {(x, y) ∈ B(u,v) ∧ B(u,v) ⊆ A} =
=
∪
(u,v)∈A
{B(u,v)|B(u,v) ⊆ A} (9)
A • B = (A ⊕ B) ⊖ BS
=
∩
(u,v)∈AC
{BC
(u,v)|B(u,v) ⊆ AC
} = (AC
◦ BS
)C
(10)
where A is a 2-D binary image object and B is a 2-D structuring element
defined on an Euclidian space, E ∈ Z2
. BS
denotes the symmetric structuring
element B. B(x,y) is the translation of B by vectors (u, v). AC
is the com-
plement of A relative to E, AC
= {(x, y)|(x, y) ∈ E ∧ (x, y) ̸∈ A}. Besides
morphological gradient (∇B) we can also define external contour (βext
B ) and
internal contour (βint
B ) for the 2-D binary images:

284 Eugen Zaharescu
∇B(A) = (A ⊕ B) − (A ⊖ B) (11)
βext
B (A) = (A ⊕ B) − A; βint
B (A) = A − (A ⊖ B) (12)
Furthermore we can define a couple of more complex MM operators as the
Top-Hat transforms. The White Top-Hat transform can detect and extract
bright structures on non-uniform backgrounds and is defined as the difference
between the input image and its opening by some structuring element:
τW
g (f)(x) = f(x) − γg(f)(x) = f(x) − δ˘g(εg(f))(x) =
= f(x) − ((f ⊖ g) ⊕ g)(x), ∀x ∈ Ef ⊆ Rn
(13)
The Black Top-Hat is the dual transform and it can detect and extract
dark structures on non-uniform backgrounds and is defined as the difference
between the closing of an image by some structuring element and the same
input image:
τB
g (f)(x) = ϕg(f)(x) − f(x) = ε˘g(δg(f))(x) − f(x) =
= ((f ⊕ g) ⊖ g)(x) − f(x), ∀x ∈ Ef ⊆ Rn
(14)
The skeleton of the input image f(x) represents the union of all intersec-
tions of the differences between the erosion and its opening through variable
sized structuring elements g(x) (i.e. White Top-Hat transforms) [7]:
S(f) =
∪
λ≥0
∩
µ≥0
[ελg(f)(x) − γµg(ελg(f)(x))], ∀x ∈ Ef ⊆ Rn
. (15)
The Hit-or-Miss transform, locates either pixels from erosion X − E1 (i.e.
”hit”) and from erosion Xc
− Ec
2, (i.e. ”miss”) [7]:
HMTE(X) = εE1 (X)
∩
εE2 (XC
) (16)
2 Morphological signatures
Morphological signatures can be obtained from a series of successive open-
ings or closings. They can be extended from a pixel-related scale to an image-
related scale (i.e. from a local definition using morphological profile operators
to a global definition using pattern spectrum/granulometry operators) ([4],
[5]).
The opening morphological signature series is defined as:
Γ(f)(x) = {γkg(f)(x)|kg = δk
S(g)(x), ∀k ∈ {0, ..., m}, ∀x ∈ Ef ⊆ Rn
} (17)

where γkg(f)(x) is the opening by a variable sized structuring element kg
and δk
S(g)(x) is the k-times successive dilation of the structuring element g
by another very simple or elementary structuring element S. By definition
γ0(f)(x) = f(x).
The dual closing morphological signature series is defined as:
Φ(f)(x) = {ϕkg(f)(x)|kg = δk
S(g)(x), ∀k ∈ {0, ..., m}, ∀x ∈ Ef ⊆ Rn
} (18)
Where ϕkg(f)(x) is the closing by a similar variable sized structuring ele-
ment kg, as defined above, and δk
S(g)(x) is the same k-times successive dilation
of the structuring element g by another very simple or elementary structuring
element S.
Subsequently, we can define the differential series computed from opening
morphological signature series, which can provide more meaningful morpho-
logical information:
∆γ(f)(x) = {γkg(f)(x) − γ(k−1)g(f)(x)|kg = δk
S(g)(x),
∀k ∈ {0, ..., m}, ∀x ∈ Ef ⊆ Rn
} (19)
The dual differential series computed from closing morphological signature
series can be defined as:
∆ϕ(f)(x) = {ϕ(k−1)g(f)(x) − ϕkg(f)(x)|kg = δk
S(g)(x),
∀k ∈ {0, ..., m}, ∀x ∈ Ef ⊆ Rn
} (20)
Finally, morphological signatures are defined from these 4 series:



Γ(f)(x)
Φ(f)(x))
∆γ(f)(x)
∆ϕ(f)(x))
(21)
where x ∈ Ef ⊆ Rn
.
On a local pixel scale this set of 4 series is used to define differential mor-
phological profile for a given pixel x ∈ Ef ⊆ Rn
in image f. On a global image
scale the pattern spectrum is built from the image series above gathering the
image pixel values through the sum operation, i.e.



∑
x∈Ef
(Γ(f)(x) + Φ(f)(x))
∑
x∈Ef
(∆γ(f)(x) + ∆ϕ(f)(x))
(22)
.

286 Eugen Zaharescu
The shape probability distribution function, involves a normalization by the
initial image volume:



∑
x∈Ef
(Γ(f)(x) + Φ(f)(x))
∑
x∈Ef
(f)(x)
∑
x∈Ef
(∆γ(f)(x) + ∆ϕ(f)(x))
∑
x∈Ef
(f)(x)
(23)
.
3 Morphological image indexing based on edge extrac-
tion and homotopic skeleton
One of the simplest morphological contour extraction methods is the morpho-
logical gradient: basically it consists of constructing an edge intensity map of
the image as the difference between the local dilation and the local erosion at
each image pixel (the word ”local” being induced by the use of a structuring
element with finite spatial support).
∇B(f)(x) = (f ⊕ g)(x) − (f ⊖ g)(x), ∀x ∈ Ef ⊆ R2
(24)
Starting from the above definition of morphological gradient for the 2-D
images we can compute the length of an object boundary or contour which is
a meaningful local morphological signature. This can also be extended to a
global image scale:
L(∇B(f)) =
∑
x∈Ef
∇B(f)(x) (25)
where x ∈ Ef ⊆ R2
.
The first step is to generate an edge intensity map using the morphological
gradient operator. The edge intensity map consists of values proportional to
the local variation within each pixel neighborhood, as defined by the support
of the structuring element. The binary edge map is obtained by thresholding
the edge intensity map and selecting the pixels with a strong color (value)
variations, measured by important values of the morphological gradient.
A typical edge extraction result by the described method is presented in
figure 1.
Also, we can compute another important morphological signature as the
length of the skeleton:

L(S(f)) =
∑
x∈Ef
∪
λ≥0
[ελg(f)(x) − γλg(ελg(f)(x))], ∀x ∈ Ef ⊆ R2
The length of the skeleton was used as a very meaningful morphological
signature in the experiment concerning labelling and classiﬁcation of color
image objects as shown in Figure 2.
Acknowledgement: The publication of this article was supported by the
grant PN-II-ID-WE-2012-4-169 of the Workshop ”A new approach in theoret-
ical and applied methods in algebra and analysis”.

288 Eugen Zaharescu
References
[1] J. Angulo, J. Serra, ”Color segmentation by ordered mergings”, Proc.
of IEEE International Conference on Image Processing (ICIP’03), IEEE,
Vol. 2, Barcelona, Spain, Sept. 2003, p. 125-128.
[2] J. Angulo, J. Serra, ”Mathematical morphology in color spaces applied to
the analysis of cartographic images”, Proc. of IEEE International Con-
ference on Image Processing (ICIP’07), 2007.
[3] Aptoula, E., Lefèvre, S.: ”A comparative study on multivariate mathe-
matical morphology”, Pattern Recognition (2007).
[4] Lefèvre, S., Weber, J., Sheeren, D.: ”Automatic building extraction in
vhr images using advanced morphological operators”, IEEE/ISPRS Joint
Workshop on Remote Sensing and Data Fusion over Urban Areas, Paris,
France (2007).
[5] Lefèvre, S., ”Extending Morphological Signature for Visual Pattern
Recognition”, Pattern Recognition (2009).
[6] J. Serra, ”Image Analysis and Mathematical Morphology”. Vol I, and
”Image Analysis and Mathematical Morphology.Theoretical Advances”.
Vol II: London: Academic Press, 1982,1988.
[7] J. Serra, ”Image Analysis and Mathematical Morphology” Academic
Press, London, 1982.
[8] P. Soille, ”Morphological Image Analysis”, Springer-Verlag, Berlin, Hei-
delberg, New York, 1999.
[9] E. Zaharescu, M. Zamfir, C.Vertan: ”Color Morphology-like Operators
based on Color Geometric Shape Characteristics”, Proc. of International
Symposium on Signals Circuits and Systems SCS 2003, Ia¸si, Romania,
2003.
[10] E. Zaharescu, ”Morphological Algorithm for Color Objects Classifica-
tion”, IEEE Computer Society Proceedings of the Fourth Balkan Confer-
ence in Informatics, 17-19 September 2009, Thessaloniki, Greece, pag.152-
156, Library of Congress No. 2009934322, (ISBN 978-0-7695-3783-2).
“Ovidius” University of Constanta,
Faculty of Mathematics and Informatics,
124 Mamaia Blvd., 900527 Constanta, Romania,
e-mail: ezaharescu@univ-ovidius.ro

Zaharescu__Eugen Color Image Indexing Using Mathematical Morphology

More Related Content

What's hot (18)

Viewers also liked (16)

Similar to Zaharescu__Eugen Color Image Indexing Using Mathematical Morphology (20)

More from Eugen Zaharescu (11)

Zaharescu__Eugen Color Image Indexing Using Mathematical Morphology