Elsevier

Knowledge-Based Systems

Volume 54, December 2013, Pages 163-171
Knowledge-Based Systems

Topological interpretations of fuzzy subsets. A unified approach for fuzzy thresholding algorithms

https://doi.org/10.1016/j.knosys.2013.09.008Get rights and content

Abstract

We show that the classical definition of a fuzzy subset carries additional structures of a topological nature. We look at the concept of a fuzzy subset and its corresponding α-cuts from an alternative point of view: namely, a fuzzy subset may be interpreted as a nested topology on a crisp set of reference, called a universe. Several kinds of fuzzy subsets associated to this interpretation are analyzed. Other topologies induced by fuzzy subsets are considered, paying special attention to their relationship with total preorders defined on the universe. Moreover, this theoretical approach allows us to provide a unified framework for most of the fuzzy thresholding algorithms that can be found in the literature.

Introduction

A new look at the classical definition of a fuzzy subset, introduced by Zadeh in his seminal work [59] (see also [9]), reveals topological and other structures that were hidden in the original definition.

Let us consider the standard definition (see [59], [9]) of a fuzzy subset1 X (of a set U) as a function μX: U  [0, 1], where U denotes a nonempty (crisp) set of reference, usually called universe. This approach is obviously equivalent to the definition of a fuzzy subset as a nested family {Uα: α  [0, 1]} of crisp subsets of U, called the α-cuts. Here Uα = {t  U: μX(t)  α}, for each α  [0, 1].

But a nested family of subsets of a given crisp set U immediately induces a topology that is also nested. Thus, the classical definition of a fuzzy subset could be (re)-interpreted as a nested topology of a certain kind defined on a given crisp set. Other remarkable topological interpretations of fuzzy sets and related items have already appeared in the literature (see e.g. [1], [31], [39]).

Moreover, different kinds of fuzzy subsets would lead to different sorts of nested topologies. To put an easy example, observe that when X = U, so that X is crisp and μX(t) = 1 for every t  U, we have that Uα = U for every α  [0, 1]. It is then obvious that the topology induced by the subbasis {Uα = U}α∈[0,1] is the trivial one on the set U. Conversely, it straightforward to see that fuzzy subsets that are not crisp induce non-trivial topologies. Note also that when X = ∅, i.e. μX(t) = 0 for each t  U, the induced topology is also the trivial one. But, for any other fuzzy set, the topology is not trivial (included the crisp sets X  U).

Which additional information is furnished by this new setting?

On the one hand, there are important classes of topologies (e.g., the lower preorderable ones) that are, by definition, nested. Also, a key fact related to nested topologies is that of defining different kinds of orderings (e.g. linear orders, total preorders) on a given nonempty crisp set. Having this in mind, we wonder if those orderings can directly be defined by means of suitable fuzzy subsets.

Moreover, on the other hand, this theoretical framework allows us to provide a general description for the most widely used fuzzy thresholding algorithms, such as those of Otsu [43], Huang and Wang [35] or the algorithm of the areas, which in some cases can be shown to be equivalent to Otsu’s algorithm [17]. Moreover, although we do not develop it here, these developments admit a straight extension to those fuzzy image thresholding algorithms based on the use of extensions of fuzzy sets, as it is the case for Tizhoosh’s algorithm [53].

The structure of the paper goes as follows:

After the Introduction (Section 1), the key notions of a fuzzy subset, a nested family and a nested topology are introduced in Section 2 of Preliminaries. In Section 3 we analyze the relationship between fuzzy subsets and nested topologies on a crisp set. In Section 4 we study, in terms of fuzzy subsets, particular classes of topologies, mainly related to the collection of total preorders that could be defined on a crisp set. In Section 5 we show an application of our theoretical developments to provide a unified framework for fuzzy image thresholding algorithms. Section 6 (Conclusion) closes the paper.

Section snippets

The classical concept of a fuzzy subset

Throughout the paper, a set in the standard sense is said to be crisp, in opposition to the term “fuzzy” introduced in Definition 2.1.

The standard definition of a fuzzy subset follows now.

Definition 2.1

[59]

Let U be a nonempty set, usually called universe. A fuzzy subset X of U is defined by means of a map μX: U  [0, 1]. The map μX is said to be the membership function (or indicator) of X.

The support of X is the crisp subset Supp(X) = {t  U: μX(t)  0}  U, whereas the kernel of X is the crisp subset Ker(X) = {t  U: μX(t) = 1} 

Nested families as regards [0, 1] vs. fuzzy subsets

As commented in Remark 2.4, a fuzzy subset generates a family of subsets of a crisp set, that is nested as regards [0, 1] in the sense of Definition 2.8.

What about the converse?

We wonder if any nested family as regards [0, 1] can be identified to a fuzzy subset, by agreeing with its collection of α-cuts. The answer is negative as Example 3.1 proves. Fortunately, we can characterize nested families as regards [0, 1] that correspond to the α-cuts of a fuzzy subset, as stated in Proposition 3.2.

Example 3.1

Let U=

From general nested families to total preorders

In Proposition 2.19 we have seen that total preorders induce nested families. But the converse is also true: as a matter of fact, a nested family O, defined on a set U, immediately endows U with a total preorder O, as Proposition 4.2 shows. First we introduce a preparatory definition.

Definition 4.1

Let U denote a nonempty crisp set. Given a nested family O={Oα}αA of subsets of U, the relation O given by aObαA(bOαaOα)(a,bU) is said to be the natural preorder associated to the nested family O.

Let us

A unified framework for fuzzy thresholding

A key factor for vision systems is the identification of regions (which represent objects) on an image. This operation, which is very simple for humans, is very difficult for machines. The division of an image into regions is known as segmentation; i.e., the segmentation of digital images is the process of dividing an image into disjoint parts, regions or classes so that each one of them has similar attributes or properties. Each of these classes represents an object of the image. There exist

Conclusion

The interpretation of several kinds of nested topologies, as well as total preorders on a nonempty crisp set U called universe, as suitable fuzzy subsets X of U carries a large variety of unexplored and maybe astonishing possibilities. Indeed, using classical well-known operators arising in fuzzy set theory (i.e., triangular norms and conorms, ordered weighted averaging aggregation operators [57], etc.) we may define operations directly on total preorders or nested topologies. The procedure is

Acknowledgement

This work is supported by the research projects MTM2009-12872-C02-02, TIN2010-15055 and MTM2012–37894-C02-02 (Spain).

Thanks are given to the editor in chief (Hamido Fujita Ph.D.), as well as to two anonymous referees, for their valuable suggestions and comments that have led to a substantial improvement of the manuscript.

References (61)

  • H. Bustince et al.

    Interval-valued fuzzy sets constructed from matrices: application to edge detection

    Fuzzy Sets and Systems

    (2009)
  • B. Caldairou et al.

    A non-local fuzzy segmentation method: application to brain MRI

    Pattern Recognition

    (2011)
  • M.J. Campión et al.

    Representability of binary relations through fuzzy numbers

    Fuzzy Sets and Systems

    (2006)
  • M.J. Campión et al.

    Preorderable topologies and order-representability of topological spaces

    Topology and its Applications

    (2009)
  • J.C. Candeal et al.

    Order representability in groups and vector spaces

    Expositiones Mathematicae

    (2012)
  • F. Feng et al.

    Soft sets and soft rough sets

    Information Sciences

    (2011)
  • J.A. Goguen

    L-fuzzy sets

    Journal of Mathematical Analysis and Applications

    (1967)
  • G. Herden

    On the existence of utility functions

    Mathematical Social Sciences

    (1989)
  • L.K. Huang et al.

    Image thresholding by minimizing the measure of fuzziness

    Pattern Recognition

    (1995)
  • V. Knoblauch

    Lexicographic orders and preference representation

    Journal of Mathematical Economics

    (2000)
  • V. Knoblauch

    Continuous lexicographic preferences

    Journal of Mathematical Economics

    (2005)
  • J. Liu et al.

    Topological analysis of knowledge maps

    Knowledge-Based Systems

    (2012)
  • P. Melo-Pinto et al.

    Image segmentation using Atanassov’s intuitionistic fuzzy sets

    Expert Systems with Applications

    (2013)
  • J. Sanz et al.

    Improving the performance of fuzzy rule-based classification systems with interval-valued fuzzy sets and genetic amplitude tuning

    Information Sciences

    (2010)
  • J. Sanz et al.

    A genetic tuning to improve the performance of fuzzy rule-based classification systems with interval-valued fuzzy sets: degree of ignorance and lateral position

    International Journal of Approximate Reasoning

    (2011)
  • M. Shabir et al.

    Another approach to soft rough sets

    Knowledge Based Systems

    (2013)
  • H. Tizhoosh

    Image thresholding using type II fuzzy sets

    Pattern Recognition

    (2005)
  • J. Ye

    Expected value method for intuitionistic trapezoidal fuzzy multicriteria decision-making problems

    Expert Systems with Applications

    (2011)
  • L.A. Zadeh

    Fuzzy Sets

    Information and Control

    (1965)
  • Z. Zhang

    On interval type-2 rough fuzzy sets

    Knowledge-Based Systems

    (2012)
  • Cited by (7)

    View all citing articles on Scopus
    View full text