Concept of linguistic variable-based fuzzy ensemble approach: application to interlaced HDTV sequences

  • Authors:

  • Gwanggil Jeon;Marco Anisetti;Joohyun Lee;Valerio Bellandi;Ernesto Damiani;Jechang Jeong

  • Affiliations:

  • School of Information Technology and Engineering, University of Ottawa, Ottawa, ON, Canada and Department of Electronics and Computer Engineering, Hanyang University, Seoul, Korea;Department of Information Technology, University of Milan, Crema, Italy;Department of Electronics and Computer Engineering, Hanyang University, Seoul, Korea;Department of Information Technology, University of Milan, Crema, Italy;Department of Information Technology, University of Milan, Crema, Italy;Department of Electronics and Computer Engineering, Hanyang University, Seoul, Korea

  • Venue:
  • IEEE Transactions on Fuzzy Systems
  • Year:
  • 2009

Quantified Score

Hi-index 0.00

Visualization

Abstract

This paper addresses the problem of edge restoration in digital images. Taking advantage of an ensemble approach, multiple type-1 fuzzy filters are combined to reach a decision. The fuzzy logic concept for linguistic variables and possibility theory is discussed with regard to knowledge representation and inference procedures. To improve conventional deinterlacing issues,we adopt type-1 fuzzy set concepts to design a weight-measuring approach. We demonstrate that the fuzzy ensemble approach model is well suited to image processing and provide case studies in the videodeinterlacing field. In our proposedmethod, five fuzzy membership functions (MFs) of linguistic variable-based fuzzy logic filters are derived from the type-1 (a.k.a. ordinary or primary) fuzzy MF. The weight-measuring process of our proposed model is used to assign weights to six candidate deinterlaced pixels (CDPs) that are interpolated according to edge direction. The use of a different MF for each direction allows the filter to characterize each pixel variation influence independently, according to its direction. The weights from all MFs are multiplied with the CDPs. The results of the empirical trials clearly show that the proposed system can successfully deal with several image types containing motion or detail elements.