Equivalence of chain codes

Hermilo Snchez-Cruz
Hiram H. Lpez-Valdez

Journal of Electronic Imaging 23(1), 013031 (JanFeb 2014)

Equivalence of chain codes

Hermilo Snchez-Cruz* and Hiram H. Lpez-Valdez
Centro de Ciencias Bsicas, Departamento de Ciencias de la Computacin, Universidad Autnoma de Aguascalientes, Av. Universidad 940,
Col. Cd Universitaria, CP. 20131 Aguascalientes, Mxico

Abstract. The representation of images is an active and very important area in image processing and pattern
recognition. Therefore, in the literature, different contour codes for binary images have been proposed, such as
F 4; F 8; V CC; 3OT , and AF 8. These codes have been used in many papers since the first chain code, F 8, was
introduced by Freeman in 1961. All the codes have been tried here as vector representations, including vertex
chain code (V CC). To know their properties, this paper provides an analysis of comparisons of each code, and
as an important contribution, we investigated the relationship between them and found a series of transformations that allow simple and efficient sets to go from one code to another. We found the equivalences between
F 4; V CC; 3OT ; F 8, and AF 8. As an important consequence of the transition matrix concept, we proposed a new
code for eight connectivity by observing a missing code in the state of the art and in the inspiration of the 3OT
code. 2014 SPIE and IS&T [DOI: 10.1117/1.JEI.23.1.013031]
Keywords: chain codes; vertex chain code; 3OT; contour; transition matrix.
Paper 13339 received Jun. 25, 2013; revised manuscript received Dec. 19, 2013; accepted for publication Jan. 23, 2014; published
online Feb. 28, 2014.

1 Introduction
Using chain codes is a common way for binary object-shape
representation. The Freeman code, proposed in 1961,1 is
known as Freeman chain code, and it is composed of
eight directions, so we call it here F8 code. It is a compact
way to represent binary object shapes, and it has the characteristic of going through a sequence of pixels of the contour on the basis of eight connectivity. Each movement
direction is codified using a symbol , where
0; 1; 2; : : : ; 7 in clockwise direction [see Fig. 1(b)].
Analogous to F8, F4 code visits the pixels of the contour
using four connectivity. A movement direction is codified
using a symbol , with 0; 1; 2; 3 [see Fig. 1(a)]. F4 is
also known as crack code because it covers the contour
shape along edges of border pixels.26
Since Freeman proposed the first chain code,1 a considerable amount of papers using chain codes for a wide variety
of issues have appeared.
One motivation to study chain codes is that, despite the
different chain codes that have been proposed independently,
a theoretical analysis is necessary to know if the codes are
related. Of course, implementing the codes takes some technical effort. For example, 3OT is more difficult to implement
than vertex chain code (VCC) because the former takes into
account relative changes regarding a reference vector, which
remains the same unless another change occurs. As can be
studied from the literature, different proposals have been
implemented independently. However, a question arises:
can a general scheme be constructed that relates the different
codes and predicts the existence of more?
Another motivation to study chain codes is that they can
be used particularly in compression of scanned documents.
For example, recently 3OT code has been utilized to compress image text documents, obtaining more than twice the
*Address all correspondence to: Hermilo Snchez-Cruz, E-mail: hsanchez@
correo.uaa.mx

Journal of Electronic Imaging

levels of compression regarding the international standard

called Joint Bilevel Image Expert Group 2.7
Other applications are, for example, for map representations,2,8 to look for dominant points,913 for analysis, and, as
we already mentioned, for document compression.7,14
Since the basic codes were appearing independently, a
question to answer is: can existing codes be obtained through
a general model? Therefore, the main objective of this paper
is to make a theoretical analysis of the different codes
existing in the literature and to find a general model that
relates them, although they emerged independently. From
this study, filling the gaps of missing codes is necessary.
Among the advantages of this study, we can say that having equivalences of the codes, we can handle different geometric interpretations of the shapes-of-objects information,
mainly depending on four or eight neighborhood in contour
representations for analysis and recognition purposes.
Depending on the model of representation used, it is necessary to work with either eight or four neighborhood. For
example, if a comprehensive analysis of discrete straight
lines detection is required, then eight neighborhood is necessary.15 On the other hand, if improving the level of compression is required, then 3OT code plays an important role.
While 3OT code improves levels of compression regarding
the VCC, the latter has some advantages in terms of representation. For example, due to its nature, VCC can give
information of concavities and convexities of the contours;
another advantage is its relationship with the Euler
number.16 Other proposals can search more geometric
properties. The codes give particular geometric interpretations depending on whether they represent four or eight
connectivity.
In 1999, Bribiesca17 proposed VCC. Among its most
important features, VCC is composed of three symbols:
1; 2; 3; it is invariant under mirror and rotation transformations and invariant to the initial point, which has some
0091-3286/2014/$25.00 2014 SPIE and IS&T

013031-1

JanFeb 2014

Vol. 23(1)

Snchez-Cruz and Lpez-Valdez: Equivalence of chain codes

Table 1 Basic codes and their derived codes.

Year proposed

Fig. 1 Freeman symbols obtained in clockwise direction: (a) F 4 code

and (b) F 8 code.

relationship between the contour and inside information.

This code represents the changes made to an object contour
by computing the number of affected pixels.
In 2005, Snchez-Cruz and Rodrguez-Dagnino18 proposed the 3OT code. They compared it with F4 and obtained
a better performance, thanks to the use of the symbols 0, 1,
and 2 to label the changes generated in relation to orthogonal
directions. 3OT code has the same number of symbols as
VCC; however, it is composed of three vectors to codify
the contour: reference, support, and change. In this case,
the symbol 0 represents no changes between reference and
support, whereas 1 represents a change equal to reference
and 2 represents a change in contrary sense of reference.
Supported on the F8, in 2005 Kui and Zalik19 proposed a
new chain code, which we call here AF8. It is based on
changes obtained with every pair of F8 code vectors
when following the contour, i.e., any vector change in the
contour is compared with the previous one, and depending
on the angle, a symbol is given. This code was compared
with F4, F8, VCC, and 3OT,20 and the results showed
that AF8 offered more advantages.
The set of codes studied is B fF4; VCC; 3OT; F8;
AF8g, and we call them basic codes (see Definition 1).
From B other codes have been derived. The derived codes
were obtained by combining the symbols that appear in the
contours and making a probabilistic model to modify the
number of bits required to store the coded contours. For
example, EVCC , V VCC , and CVCC were proposed in
Ref. 21. EVCC was obtained by considering that VCC
uses two bits to represent three symbols, V VCC arises by considering a variable length of VCC chain, and CVCC is based
on applying the Huffman algorithm on the VCC code symbols. On the other hand, M3OT was proposed by considering
grouping symbols of 3OT,22 whereas MDF9 was proposed
by considering the AF8 patterns in pieces of discrete straight
lines of the contour shapes.15 Table 1 shows the basic and the
derived codes.
According to different criteria to satisfy the need to represent the object shapes and looking for relationships of
contours and inner regions, in the literature, it is reported
that each code of the set B was obtained independently
and in different years. However, we demonstrate that there
is an equivalence of all basic codes, even more there is a
way to go from one code to another. In 1997, Wilson
found an equivalence between F8 and F4.6 However, in
this work, we found the equivalence between the basic
codes proposed since then.
In this paper, we demonstrate that there is a transition
matrix that allows us to go from one code C1 to another
Journal of Electronic Imaging

Basic code

Derived codes

1961

F8; F4

1999

V CC

E V CC , C V CC , V V CC

2005

3OT

M3OT

2005

AF8

MDF9

C2 , with Ci B. If this happens, we say the codes C1

and C2 are equivalent.
The transition matrix that allows us to go from a code to
another will be very important because if a coding has some
properties, like rotation invariance, they have to be reflected
in the transition matrix.
Of course, all the codes can be handled as vector components, VCC inclusive, as we show in this paper. The mentioned codes are useful to represent contour shapes in
either four or eight connectivity. A summary of the basic
codes for eight and four connectivity contours appears in
Table 2. According to the number of vectors used to codify,
we classify the basic codes in three types.

Type 1 is composed of the codes that use one vector.

Type 2 is composed of the codes that use two vectors.
Type 3 is composed of the codes that use three vectors.
This information is shown in Table 2, in which we notice
that there are two codes of types 1 and 2; however, there is a
missing code for type 3 represented by a small box, which
has not been proposed. In this work, after analyzing matrix
transformations to find code equivalences, we discover this
missing code.
In Sec. 2, we introduce the basis of the codes F4; VCC;
3OT; F8, and AF8. In Sec. 3, we study the case of four connectivity, and we demonstrate that F4, VCC, and 3OT are
equivalent. In Sec. 4, we deal with the case of eight connectivity, and we demonstrate that F8 and AF8 are equivalent.
Despite that some codes are for eight connectivity and others
are for four connectivity, in Sec. 5, we demonstrate that all
the codes of the basis are equivalent. As a result of equivalences of chain codes, in Sec. 6, we found a new code
for eight connectivity. In Sec. 7, we test our new code on
some sample objects and obtain some results. Finally, in
Sec. 8, we present some conclusions and future work.
2 Concepts and Definitions
The codes that we use in this work are F4; VCC; 3OT; F8,
and AF8; from these we give the following definitions:
Table 2 Bases of chain codes representing four and eight connectivity. The box represents a missing code.
One vector

Two vectors

Three vectors

4 connectivity

F4

V CC

3OT

8 connectivity

F8

AF8

013031-2

JanFeb 2014

Vol. 23(1)

Snchez-Cruz and Lpez-Valdez: Equivalence of chain codes

Definition 1 The set of codes studied is B fF4; VCC;

3OT; F8; AF8g, and we call them basic codes.
Definition 2 If a figure is codified with some code X , and
the coding is
C fCX 1; : : : ; CX ng;
then the number n is called the length of the code X .

2.1 F 4 and F 8 as a Vector Representation

Traditionally, F4 and F8 symbols are obtained by considering their basis vector in counterclockwise direction.
However, for convenience, in this paper, we use the clockwise direction (see Fig. 1).

2.2 V CC as a Vector Representation

The VCC was initially conceived as a code based on the
number of pixel vertices that are in touch with the boundary
contour of the shape, giving three cases coded by an alphabet
of three symbols. However, by analyzing the nature of this
code, we can give it a vector interpretation. As can be seen in
Fig. 2 for VCC basis, given two connected resolution pixels,
there are three adjacent edges at the vertex of the contour.
The two edges of the outer contour can be used as directed
segments in the same direction. On the other hand, when a
pixel resolution is visited and we see that the visited vertex
has two adjacent edges, we define two vectors and consider
one of them as a vector change to the right. Finally, if the
vertex visited is adjacent to three pixels of resolution,
there is a vector change to the left.
So, when we use VCC code we need to analyze every
two vectors (first and second). If the second vector has
the same direction as the first, we put 0. If the second vector

Fig. 2 Bases used for the codes, all with vector representations, including V CC code. C F 4 f0; 0; 0; 0;
1; 1; 1; 0; 1; 1; 2; 1; 2; 1; 2; 2; 2; 2; 3; 2; 3; 3; 0; 0; 0; 3; 2; 2; 3; 3; 0; 3g, C V CC f1; 0; 0; 0; 1; 0; 0; 2; 1; 0; 1; 2;
1; 2; 1; 0; 0; 0; 1; 2; 1; 0; 1; 0; 0; 2; 2; 0; 1; 0; 1; 2g, C 3OT f0; 0; 0; 2; 0; 0; 1; 1; 0; 2; 1; 1; 1; 1; 0; 0; 0; 2; 1; 1;
0; 2; 0; 0; 1; 2; 0; 1; 0; 2; 1; 1g.
Journal of Electronic Imaging

013031-3

JanFeb 2014

Vol. 23(1)

Snchez-Cruz and Lpez-Valdez: Equivalence of chain codes

changes 90 deg to the right regarding the first, we put 1. If

the second vector changes 90 deg to the left regarding the
first, we put 2. The vectorial base, or simply base, is given
in Fig. 2.
2.3 3OT as a Vector Representation
In the case of 3OT, we need to use three vectors to codify the
contour: reference, support, and change. The symbol 0 represents no changes between reference and support, 1 represents a change in the same direction to a reference, and
2 represents a change in contrary sense of the reference.
Of course, support vector can be a sequence of r vectors
in the same direction, with r 1.
Figure 3 shows an example of coding using F4, VCC,
and 3OT.
Before starting to define the bases, we mention some
features that the codes of B have in common.
Note 1 F4; VCC; 3OT; F8, and AF8 have the following
common features (Figs. 3 and 4 illustrate some
examples):

Fig. 4 The figure is covered through centers of pixels. C F 8 f0; 0;

0; 2; 2; 1; 2; 3; 3; 4; 4; 4; 5; 6; 0; 0; 7; 5; 4; 4; 6; 7g, C AF 8 f1; 0; 0; 2; 0;
7; 1; 1; 0; 1; 0; 0; 1; 1; 2; 0; 7; 6; 7; 2; 1g.

1. Contour pixels are visited on clockwise direction.

2. By simplicity, the first pixel is visited at leftmost top
of the figure.
The second part of Note 1 is a convention. We
have always started with this pixel to compare different chain codes of the same figure. Of course, another
convention is possible if starting with another pixel is
required.
Note 2 F4, VCC, and 3OT have the following common
features (Fig. 3):
1. Edges of contour pixels are visited.
2. The first vertex visited is in the leftmost part of the
top of the first pixel.
3. The first two contour edges of the first pixel visited
correspond to 3 and 0 symbols of F4 code.

Fig. 3 The figure is covered through outer vertex.

Journal of Electronic Imaging

4. The first change of a discrete curve corresponds to

the symbol 2 of 3OT code.
5. The chain codes have the same length.
Note 3 F8 and AF8 have the following common features
(see Figs. 4 and 5):
1. Centers of contour pixels are visited.
2. The first center visited is that of the first pixel.
3. The chain codes have the same length.
The following three notes are a consequence of
Table 2.
Note 4 If F4 or F8 needs to be coded/decoded, vector by
vector have to be analyzed.
Note 5 If VCC or AF8 needs to be coded/decoded, every
two consecutive vectors have to be analyzed.
Note 6 If 3OT needs to be coded/decoded, every three consecutive vectors have to be analyzed.
Note 7 As usual, F4 and F8 codes are used following clockwise direction (see Fig. 1).

Fig. 5 The figure is covered through centers of pixels. C F 8 f0;

0; 0; 2; 3; 4; 3; 5; 6; 7g; C AF 8 f1; 0; 0; 2; 1; 1; 7; 2; 1; 1g.
013031-4

JanFeb 2014

Vol. 23(1)

Snchez-Cruz and Lpez-Valdez: Equivalence of chain codes

matrix F4 T VCC , which can be read like a multiplication

table.

Figure 4 shows an example of codification using

F8 and AF8.
Definition 3 Two codes C1 and C2 are equivalent if it is
possible to go from one code to another without visiting
object shapes; in this case, we write C1 C2 .
Definition 4 Let C1 and C2 be equivalent codes. A
matrix M is called transition matrix from C1 to C2 if it allows
to go from the code C1 to the other code C2 . If this happens,
this matrix M is denoted by C1 T C2 .
3 F4, VCC, and 3OT are Equivalent
In this section, we demonstrate F4 VCC and F4 3OT.
From linear algebra, a matrix T of r rows and s columns
can be written as

Of course, this analysis can be carried out independent

of the vector orientations. So, transition matrix from
VCC to F4 VCC T F4 is

tpq 44 ;

where tpq Sqp mod 4 . Conversely, a transition matrix from

VCC to F4 is

if q f1; 2g
:
if q 3

F4 to VCC: Assume that a figure is codified with F4,

and the coding is
CF4 fCF4 1; : : : ; CF4 i; : : : ; CF4 ng:

Let us denote by CVCC the chain code we want to obtain

of the same figure coded by F4. By Note 1 and Note 2,
the first two orthogonal vectors of the first pixel compose
the first code symbol, 1, of VCC, i.e., CF4 n 3 and
CF4 1 0, and CVCC 1 1. While we cover the F4
chain, we have to focus on two contiguous vectors.
Now, taking into account every pair of vectors from
F4 basis, we can obtain one of the symbols of VCC
basis, independent of orientation. Observe that if CF4 i
1 j and CF4 i j, then CVCC i 0,
CF4 i 1 j and CF4 i j 1 mod 4, then
CVCC i 1, CF4 i 1 j and CF4 i j
2 mod 4 never happens (for this is used S2
),
if CF4 i 1 j and CF4 i j 3 mod 4, then
CVCC i 2, where i 2; ; n and j f0; 1; 2; 3g.
These facts can be represented with the transition
if

Journal of Electronic Imaging

B1
@
2
3

1
1
2
3
0

2
31
0C
A
1
2

which we can also read like a multiplication table. For

example, if we have CF4 i 1 2 and we have
CVCC i 1, then CF4 i 3.

Proof.

0
00

0
1
2
3

tpq 43 ;
p q 2 mod 4
p 2 mod 4

VCC to F4: Assume that a figure is codified with

VCC, and the coding is

if CF4 i 1 0 and CVCC i 0, then CF4 i 0,

if CF4 i 1 0 and CVCC i 1, then CF4 i 1,
if CF4 i 1 0 and CVCC i 2, then CF4 i 3.

Theorem 1. F4 VCC If S0 0, S1 1. S2
, and

S3 2, then a transition matrix from F4 to VCC is

tpq

3
21

C
A
1
0

Let us denote by CF4 the coding of the same figure with

F4. For Note 1 and Note 2, CF4 n 3 and CF4 1 0.
To find the element CF4 i, for i 2; ; n, we analyze
the elements CF4 i 1 and CVCC i. Thus, we have

where tpq represents the element of row p and column q of

the matrix T.

where

2


1
0
2

CVCC fCVCC 1; : : : ; CVCC i; : : : ; CVCC ng:

T tpq rs ;

VCC T F4

B2
@


1

1
1
0
2

For example, if we have in some place of F4: a symbol 2 followed by 3, in VCC we have 1.

F4 T VCC

0
00

0
1
2
3

Now see how to go from F4 to 3OT and vice versa.

Theorem 2. (F4 3OT) A transition matrix from F4 to
3OT is
F4 T 3OT

tpq 44 ;

where
(
tpq

1 if p q;
2 if jp qj 2 :

otherwise

And a transition matrix from 3OT to F4 is

3OT T F4

tpq 42 ;

where tpq p 1q mod 4.

013031-5

Proof.

F4 to 3OT: Assume that a figure is codified with F4,

and the coding is
JanFeb 2014

Vol. 23(1)

Snchez-Cruz and Lpez-Valdez: Equivalence of chain codes

CF4 fCF4 1; : : : ; CF4 i; : : : ; CF4 ng:

Denote by C3OT the coding of the same figure with 3OT.
For Note 1 and Note 2 CF4 n 3 and CF4 1
0. For i 1; 2; ; n, we find C3OT i with the help
of CF4 i 1, CF4 i, and CF4 k, where k < i
and CF4 k CF4 k 1 CF4 k 2 CF4 i.
If CF4 1 CF4 2 CF4 i, i.e., if there is no
k that satisfies the condition, then k n. Of course,
CF4 n 1 CF4 1. Now see the following facts:
If CF4 i 1 CF4 i, then C3OT i 0.
On the contrary, if CF4 i 1 CF4 i, we have two
options:
*
*

Theorem 3. F8 AF8 A transition matrix from F8 to

AF8 is
F8 T AF8

where tpq q p mod 8 and a transition matrix from AF8

to F8 is
AF8 T F8

Proof.

if CF4 i 1 CF4 k, then C3OT i 1,

if CF4 i1CF4 k2mod4, then C3OT i2.

0
01
B

@
2

1


1


2

2
2


1

Let us denote by CAF8 the coding of the same figure with

AF8. For i 1; 2; ; n, we find CAF8 i with the help
of CF8 i and CF8 i 1, where CF8 0 CF8 n. We
have the following facts:
if CF4 i CF4 i 1 r;

C3OT fC3OT 1; : : : ; C3OT i; : : : ; C3OT ng:

Denote by CF4 the coding of the same figure with F4. For
Note 1 and Note 2, CF4 n 3 and CF4 1 0, and to
find the element CF4 i, for i 2; ; n, the element
CF4 k (k defined in proof of F4 to 3OT) and the element
C3OT i 1 are analyzed.
If C3OT i 1 0, then CF4 i CF4 i 1,
if C3OT i 1 1, then CF4 i CF4 k,
if C3OT i 1 2, then CF4 i CF4 k 2 mod 4.

These facts can be represented by the transition matrix

T
F8 AF8 as

B7
B
B6
B
B5
B
B4
B
B3
@
2
1

1
1
0
7
6
5
4
3
2

2
2
1
0
7
6
5
4
3

3
3
2
1
0
7
6
5
4

4
4
3
2
1
0
7
6
5

5
5
4
3
2
1
0
7
6

6
6
5
4
3
2
1
0
7

7
71
6C
C
5C
C
4C
C
3C
C
2C
A
1
0

which can be read like a multiplication table. For example, if

we have in some place of F8 4 followed by 3, in AF8, we
have 7.

2
21
3C
A
0
1

and taking into account that we have C3OT i 1 0,

CF4 i CF4 i 1.

4 F8 and AF8 are Equivalent

In this section, we give a proof that F8 and AF8 are
equivalent.
Journal of Electronic Imaging

0
00

0
1
2
3
4
5
6
7

This study can be represented with the transition matrix

3OT T F4 as
0
1
2
3

with 0 r 7;

then CAF8 i r:

3OT to F4: Assume that a figure is codified with 3OT,

and the coding is

1
00
B1
@
2
3

F8 to AF8: Assume that a figure is codified with F8,

and the coding is
CF8 fCF8 1; : : : ; CF8 i; : : : ; CF8 ng:

3

1
2C
A


1

which can be read like a multiplication table. For example, if

we have in some place CF4 j 3 and CF4 i 1 1, in
3OT, we have C3OT i 2.

tpq 88 ;

where tpq p q 2 mod 8.

These facts can be represented with the transition matrix

F4 T 3OT as
0
1
2
3

tpq 88 ;

AF8 to F8: Assume that a figure is codified with AF8,

and the coding is
CAF8 fCAF8 1; : : : ; CAF8 i; : : : ; CAF8 ng:

Denote by CF8 the coding of the same figure with F8.

Because AF8 is invariant under rotation, we can suppose
that CF8 1 0. To find the element CF8 i, for i
2; ; n, we analyze the elements CF8 i 1 and
CAF8 i. Next conditions take place:
If CAF8 i r;

with 0 r 7;

then CF8 i CF8 i 1 r mod 8:

Because there are 8 8 different combinations of CAF8 i
and CF8 i 1, this study is represented by the transition

013031-6

JanFeb 2014

Vol. 23(1)

Snchez-Cruz and Lpez-Valdez: Equivalence of chain codes

matrix AF8 T F8 . The transition matrix can be read like a multiplication table.

0
1
2
3
4
5
6
7

0
00
B1
B
B2
B
B3
B
B4
B
B5
@
6
7

1
1
2
3
4
5
6
7
0

2
2
3
4
5
6
7
0
1

3
3
4
5
6
7
0
1
2

4
4
5
6
7
0
1
2
3

5
5
6
7
0
1
2
3
4

6
6
7
0
1
2
3
4
5

7
71
0C
C
1C
C
2C
C
3C
C
4C
A
5
6

Theorem 4. (F4 F8)

F4 T F8

For example, if we have CAF8 i 3, and CF8 i 1

5, then CF8 i 0.
Example 1 In this example, we show how to use the transition matrix F8 T FA8 . The chain code CF8 of Fig. 5 is
CF8 f0; 0; 0; 2; 3; 4; 3; 5; 6; 7g, and the transition matrix
F8 T FA8 can be read like a multiplication table.
0
1
2
3
4
5
6
7

0
00
B7
B
B6
B
B5
B
B4
B
B3
@
2
1

1
1
0
7
6
5
4
3
2

2
2
1
0
7
6
5
4
3

3
3
2
1
0
7
6
5
4

4
4
3
2
1
0
7
6
5

5
5
4
3
2
1
0
7
6

6
6
5
4
3
2
1
0
7

5 All Chain Codes are Equivalent

The coded vectors of F4 visit outer edges of the contours,
whereas those of F8 visit the centers of pixels, and also
F4 is a four connected representation, whereas F8 is for
eight connectivity. In this section, we prove one of the
main results of this paper: F4 VCC 3OT F8 AF8.

7
71
6C
C
5C
C
4C
C
3C
C
2C
A
1
0

2
3

4
5

71

C
A

and

0 0 010 01 0121





03 1
1 121 12




3 C
B 10
C
B
10
1 121 12 1232


C
B

C
B
0
21
2 232 23

C
B

T

C:
B
F8 F4




21
2 232 23 2303 C
B

C
B




1
32
3 303 C
B 30
A
@
30 2010





32
3 303
0 010 01




2
03
Proof.

F4 to F8: Let us assume that a figure is codified with

F4, and the coding is
CF4 fCF4 1; : : : ; CF4 i; : : : ; CF4 ng:

Taking into account two consecutive symbols of Fig. 5, we

have the following steps:
CF8 0 7 and CF8 1 0, then CAF8 1 1 [position
(7,0) of the table]
CF8 1 0 and CF8 2 0, then CAF8 2 0 [position
(0,0) of the table]
CF8 2 0 and CF8 3 0, then CAF8 3 0 [position
(0,0) of the table]
CF8 3 0 and CF8 4 2, then CAF8 4 2 [position
(0,2) of the table]
CF8 4 2 and CF8 5 3, then CAF8 5 1 [position
(2,3) of the table]
CF8 5 3 and CF8 6 4, then CAF8 6 1 [position
(3,4) of the table]
CF8 6 4 and CF8 7 3, then CAF8 7 7 [position
(4,3) of the table]
CF8 7 3 and CF8 8 5, then CAF8 8 2 [position
(3,5) of the table]
CF8 8 5 and CF8 9 6, then CAF8 9 1 [position
(5,6) of the table]
CF8 9 6 and CF8 10 7, then CAF8 10 1
[position (6,7) of the table]
So
CAF8 f1; 0; 0; 2; 1; 1; 7; 2; 1; 1g:

Let CF8 be the coding of the same figure with F8. For
i 1; 2; ; n, we find CF8 i with the help of CF4 i
and CF4 i 1, where CF4 0 CF4 n. We have to
analyze every two vectors in F4 and see how they become
in F8. Next conditions take place.
If CF4 i 1 0 and CF4 i 0, then CF8 i 0,
if CF4 i 1 0 and CF4 i 1, then CF8 i
empty, since for F8 we are in the center of the
same pixel, so we do nothing.
CF4 i 1 0 and CF4 i 2 never happens (for this

is used),
if CF4 i 1 0 and CF4 i 3, then CF8 i 7,
where i 1; : : : ; n. All these facts can be summarized
in the next transition matrix.
0
1
2
3

0
00
B1
@

2
3

2 3

71

C:
A
4
5 6

F8 to F4: Let us assume that a figure is codified

with F8, and the coding is
CF8 fCF8 1; : : : ; CF8 i; : : : ; CF8 ng:

Denote by CF4 the coding of the same figure with F4.

For Note 1 and Note 2 CF4 n 3 and CF4 1 0.

This result can be verified geometrically in Fig. 5.

Journal of Electronic Imaging

00
B1
@

013031-7

JanFeb 2014

Vol. 23(1)

Snchez-Cruz and Lpez-Valdez: Equivalence of chain codes

It can also be read like a multiplication table. For example, if we have CF8 i 1 5 and CF8 i 3, then
CF4 i 1.

Theorem 12. F4 VCC 3OT F8 AF8

Fig. 6 More vectors are necessary in F 4 than in F 8 to code the same

set of pixels.

For i 2; ; n we find CF4 i with the help of

CF8 i and CF8 i 1. We have to analyze every
two vectors in F8 and we see how they become in
F4. Some examples are shown below.
If CF8 i 1 0 and CF8 i 0, then CF4 i 0,
if CF8 i 1 0 and CF8 i 1, then CF4 i 010
because we need the vectors 0, 1, and 0 in F4 to
visit the same pixels of vector 1 in F8 (Fig. 6),

Proof. Using Theorem 1 we have F4 VCC, and for

Theorem 2, we have F4 3OT. Thanks to Theorems 3
and 4, we know that F8 AF8 and F8 F4, respectively.
Thus, the assertion is true.

6 New Code
As an important application of the transition matrix concept,
we create a new code using a matrix. This code is inspired in
3OT, but the new code is obtained by visiting the centers of
pixels. Let us call the new code AAF8 (because it uses two
angles to determine a symbol).

CF8 i 1 0 and CF8 i 4 never happens,

CF8 fCF8 1; : : : ; CF8 i; : : : ; CF8 ng:

if CF8 i 1 0 and CF8 i 7, then CF4 i 03,

where i 2; ; n. All these facts can be written in
the following transition matrix:
0
1
2
3
4
5
6
7

F8 to AAF8: Let us assume that a figure is codified

with F8, and the code is

0
1
2
3
4
5
6
7
0 0 010 01 0121





03 1
1 121 12




3 C
B 10
C
B
10
1 121 12 1232


C
B

C
B
0
21
2 232 23

C
B

C
B




21
2 232 23 2303 C
B

C
B




1
32
3 303 C
B 30
A
@
30 2010





32
3 303
0 010 01




2
03

For i 1; 2; ; n, we create CAAF8 i with the help

of CF8 i 1; CF8 i, and CF8 k, where k < i and
CF8 k CF8 k 1 CF8 k 2 CF8 i.
If
CF8 1 CF8 2 CF8 i, kcn. It is defined that
CF8 n 1 CF8 1. Thus,
if CF8 i 1 CF8 i, then CAAF8 i 8.
On the contrary, if CF8 i 1 CF8 i, then consider
the next cases.
If CF8 i 1 CF8 k r mod 8, with 0 r 7, then
CAAF8 i r.
So, CAAF8 i arises from the following transition matrix
AF8 T AAF8 :

Fig. 7 The set of allowed vector arrays of our proposed AAF 8 code.
Journal of Electronic Imaging

013031-8

JanFeb 2014

Vol. 23(1)

Snchez-Cruz and Lpez-Valdez: Equivalence of chain codes

0
1
2
3
4
5
6
7

0
00
B7
B
B6
B
B5
B
B4
B
B3
@
2
1

1
1
0
7
6
5
4
3
2

2
2
1
0
7
6
5
4
3

3
3
2
1
0
7
6
5
4

4
4
3
2
1
0
7
6
5

5
5
4
3
2
1
0
7
6

6
6
5
4
3
2
1
0
7

7
71
6C
C
5C
C
4C
C
3C
C
2C
A
1
0

This matrix is written like that which transforms F8 to

AF8 because both show what happens between reference
and change vectors. However, in this transformation, a support vector is involved, and no matter the direction of support
vector, we consider that reference and change vectors give
the codification of the shape.
Since the new code is
CAAF8
Fig. 8 The figure is covered through the centers of pixels. C F 8
f0; 0; 0; 2; 2; 1; 2; 3; 3; 4; 4; 4; 5; 6; 0; 0; 7; 5; 4; 6; 7g, C AAF 8 f8; 8; 3;
8; 1; 0; 2; 8; 2; 8; 8; 2; 2; 3; 8; 1; 5; 5; 1; 3; 2g.

fCF8 n;CF8 1; CAAF8 3;::: ;CAAF8 i;:: :;CAAF8 ng;

where CAAF8 1 CF8 n and CAAF8 2 CF8 1, we
should keep the information of CF8 n and CF8 1 to find
the transition matrix AAF8 T F8 . As can be observed, given
the first two known F8 vectors, a third vector (the change

Fig. 9 Images encoded using F 4; V CC; 3OT ; F 8; AF 8, and AAF 8.

Journal of Electronic Imaging

013031-9

JanFeb 2014

Vol. 23(1)

Snchez-Cruz and Lpez-Valdez: Equivalence of chain codes

vector) composes one of the symbols given in Fig. 7, which

shows a graphical aspect of the different symbols of the
new code.
A very important observation arises: to consider the first
change of direction, it is necessary to know the first two vectors F8n and F81, reference and support, respectively, so
that the third vector (of the change) is oriented with a certain
angle with respect to the reference, and codify the change
with one of the eight symbols of Fig. 7 (clearly, symbol 8
actually does not represent change of direction). In the following change to encode, the reference vector becomes in
what was of support F81 and the new vector of change
has a certain angle regarding reference vector. The behavior
of the support vector was already coded in the previous step,
so the information of the actual path of the shape is
never lost.

Table 3 Original sizes, number of pixels, and lengths of the different

codes.
Object

Size

Number of pixels

l4

l8

Bull

286 278

38,115

2122

1651

Map

310 213

21,747

1560

1120

Giraffe

197 505

36,945

2866

2316

Frog

183 151

12,789

1704

1172

Omega

187 184

14,565

1492

1178

VW

447 204

57,337

1388

1058

Bat

550 265

53,096

2054

1444

Motorcycle

853 695

289,672

5044

3689

Camel

293 285

32,202

1796

1330

Eagle

226 213

19,363

1214

816

Let CF8 be the coding of the same figure with F8. To find
the element CF8 i, for i 2; ; n 1, consider, again,
k < i and the condition that CF8 n and CF8 1 are known.

Dolphin

305 284

27,613

1394

1022

Pitbull

220 246

26,126

1328

1019

If CAAF8 i r;

Trex

431 287

34,053

2208

1651

Horse

238 174

11,886

1836

1277

Boot

269 359

45,787

1458

1144

Car

309 204

47,973

1202

1010

Star

259 244

28,244

1178

860

Chicken

184 180

17,946

966

678

Jump

203 244

13,776

1378

947

Hand

309 287

47,132

1864

1371

Penguin

122 160

8500

831

564

Car 2

406 221

1085

1378

1179

Dog

230 275

23,805

1562

1152

Fish

221 171

22,037

1158

841

Spiral

187 181

14,501

3128

2213

Guitar

24 77

584

240

186

AAF8 to F8: Assume that a figure is codified with

AAF8, and the code is

CAAF8
fCF8 n;CF8 1;CAAF8 3;:::;CAAF8 i;:::;CAAF8 ng:

with 0 r 7;

then CF8 i 1 CF8 k r mod 8:

Of course, there are also eight different values for CF8 k,
so there is a complete set of 64 combinations of CAAF8 i and
CF8 k to obtain a symbol CF8 i 1, which can be written
in terms of the next table.
0
1
2
3
4
5
6
7

0
00
B1
B
B2
B
B3
B
B4
B
B5
@
6
7

1
1
2
3
4
5
6
7
0

2
2
3
4
5
6
7
0
1

3
3
4
5
6
7
0
1
2

4
4
5
6
7
0
1
2
3

5
5
6
7
0
1
2
3
4

6
6
7
0
1
2
3
4
5

7
71
0C
C
1C
C
2C
C
3C
C
4C
A
5
6

Finally, if we have CAAF8 i 8, then CF8 i 1

CF8 i.
Each row of the matrix represents CAAF8 symbol, whereas
the columns represent CF8 .
The complete set of vector arrays represented by AAF8
symbols appear in Fig. 7.
The difference of this equivalence with that of F8 to AF8
is that in this a support vector is involved, and this vector can
have one of the eight different directions of F8 code.
Figure 8 shows an example of codification using F8
and AAF8.
To assign a symbol to each change direction in Fig. 8, we
can be assisted by Fig. 7. Observe that we can choose the
vector patterns that best match the contour in the position
we are coding depending on the angle given mainly by
Journal of Electronic Imaging

Trice

356 164

25,758

1460

1056

Plane

355 109

13,306

1216

932

T-shirt

191 206

23,890

860

682

Jeans

220 297

35,541

1410

1121

013031-10

JanFeb 2014

Vol. 23(1)

Snchez-Cruz and Lpez-Valdez: Equivalence of chain codes

the change direction vector regarding the reference vector

and taking into account if the represented vector pattern
of the contour points to the right or to the left of the direction
of travel.
7 Evaluation and Results
We tested our method on a sample of shapes. We evaluated
the different codes over 30 different irregular shapes that
appear in Fig. 9, obtained from a repository,23 whose
sizes and number of pixels are given in Table 3. The images
shown in Fig. 9 were coded using F4; VCC; 3OT; F8; AF8,
and AAF8.
Let l4 be the length (i.e., the number of symbols in the
chains) of the F4, VCC, and 3OT chain codes, and l8 the
length of F8, AF8, and AAF8 chains. Table 3 also presents
the lengths of the chain codes.
A transition matrix makes it possible to go from any chain
code to another independently if it is to represent four or
eight neighborhood.
If the basic codes are equivalent, they have different information content. However, since there is a conflict between
the code representation in four or eight neighborhood
when recognition tasks are carried out, the feasibility in
the representation should be taken into account when looking for patterns within chains.
8 Conclusions and Further Work
We have obtained a new relative code composed of three vectors: a reference, support, and change direction vector for
eight connectivity.
A method to obtain the main codes of the literature has
been developed. This method allows us to go from one code
to another. So we have demonstrated that all basic codes
are equivalent, although each of them was proposed
independently.
We have observed a missing code, of three vectors, to
complete the schema. Such a code is composed of three vectors for eight connectivity, which we have called AAF8. This
new code is relative and two angles are involved in its
construction.
The main characteristics of AAF8 are as follows: it is
invariant under rotation transformation and as the other
eight connected codes, AAF8 has the smallest length chain.
Just as with previous codes, future works should be
implemented for analysis and recognition tasks and, also,
to find and compare derived codes from AAF8.
Acknowledgments
We would like to thank Universidad Autnoma de
Aguascalientes under grant PIINF14-2.
References
1. H. Freeman, On the encoding of arbitrary geometric configurations,
IRE Trans. Electron. Comput. EC-10(2), 260268 (1961).
2. A. Akimov, A. Kolesnikov, and P. Franti, Lossless compression of map
contours by context tree modeling of chain codes, Pattern Recognit.
40(3), 944952 (2007).

Journal of Electronic Imaging

3. S. Blackburn, Extraction of color region boundaries, in IAPR

Workshop on Machine Vision Applications, pp. 6366, MVA'92,
Tokyo, Japan (1992).
4. L. Cederberg, Chain-link coding and segmentation for raster scan
devices, Comput. Graphics Image Process. 10(3), 224234 (1979).
5. T. Pavlidis, Algorithms for Graphics and Image Processing, Computer
Science Press, Rockville, MD (1982).
6. G. Wilson, Properties of contour codes, IEE Proc. Vis., Image Signal
Process. 144(3), 145149 (1997).
7. H. Snchez-Cruz and M. A. Rodrguez-Daz, Binary document image
compression using a three-symbol grouped code dictionary, J.
Electron. Imaging 21(2) 023013 (2012).
8. A. J. Pinho, Adaptative context-based arithmetic coding of arbitrary
contour maps, IEEE Signal Process. Lett. 8(1), 46 (2001).
9. F. Arrebola and F. Sandoval, Corner detection and curve segmentation
by multi resolution chain-code linking, Pattern Recognit. 38(10),
15961614 (2005).
10. H. Beus and S. Tiu, An improved corner detection algorithm based on
chain coded plane curves, Pattern Recognit. 20(3), 291296 (1987).
11. H. Freeman and L. S. Davis, A corner-finding algorithm for chaincoded curves, IEEE Trans. Comput. C-26(3), 297303 (1977).
12. H. Snchez-Cruz, A proposal method for corner detection with an
orthogonal three-direction chain code, in Advanced Concepts for
Intelligent Vision Systems Lecture Notes in Computer Science, J.
Blanc-Talon et al., Ed., Vol. 4179, pp. 161172, Springer Berlin
Heidelberg (2006).
13. H. Snchez-Cruz and E. Bribiesca, Polygonal approximation of contour shapes using corner detectors, J. Appl. Res. Technol. 7(3), 275
291 (2009).
14. S. Hoque, K. Sirlantzis, and M. C. Fairhust, A new chain-code quantization approach enabling high performance handwriting recognition
based on multi-classifier schemes, in IEEE Proc. of the Seventh Int.
Conf. on Document Analysis and Recognition, pp. 834838, IEEE
Computer Society, Washington, DC (2003).
15. H. Snchez-Cruz, Proposing a new code by considering pieces of discrete straight lines in contour shapes, J. Vis. Commun. Image
Represent. 21(4), 311324 (2010).
16. J. H. Sossa-Azuela et al., Computing the Euler number of a binary
image based on a vertex codification, J. Appl. Res. Technol. 11(3),
360370 (2013).
17. E. Bribiesca, A new chain code, Pattern Recognit. 32(2), 235251
(1999).
18. H. Snchez-Cruz and R. M. Rodriguez-Dagnino, Compressing bilevel
images by means of a three-bit chain code, Opt. Eng. 44(9), 097004
(2005).
19. Y. Kui-Liu and B. Zalik, An efficient chain code with Huffman coding, Pattern Recognit. 38(4), 553557 (2005).
20. H. Snchez-Cruz, E. Bribiesca, and R. M. Rodriguez-Dagnino,
Efficiency of chain codes to represent binary objects, Pattern
Recognit. 40(6), 16601674 (2007).
21. Y. Kui-Liu et al., Compressed vertex chain codes, Pattern Recognit.
40(11), 29082913 (2007).
22. H. Snchez-Cruz, M. Lpez-Cruces, and H. Puga, A proposal modification of the 3OT chain code, in Proc. of the 10th IASTED Int. Conf.
on Computer Graphics and Imaging, pp. 611, ACTA Press, Anaheim,
CA (2008).
23. Images, Bilevel_images.zip, https://sites.google.com/site/herssan/
images.
Hermilo Snchez-Cruz received his PhD in computer science in
2002 from the Universidad Nacional Autnoma de Mxico (UNAM).
Also, he was an assistant researcher with the Instituto de
Investigaciones en Matemticas Aplicadas y en Sistemas at
UNAM in 1998. Currently, he is a full-time professor with the
Universidad Autnoma de Aguascalientes in Mxico (UAA). His
areas of interest are pattern recognition, image compression, bidimensional and three-dimensional (3-D) image recognition.
Hiram H. Lpez-Valdez received his MSc in mathematics in 2010
from Centro de Investigacion y de Estudios Avanzados del
Instituto Politcnico Nacional (CINVESTAV-IPN). Currently, he is a
PhD candidate, also from CINVESTAV-IPN. He is a professor at
UAA, where he teaches undergraduate courses in image processing,
pattern recognition, and mathematics. He is interested in solving problems of features selection of objects in images, as well as topology
aspects of two-dimensional and 3-D objects.

013031-11

JanFeb 2014

Vol. 23(1)