Finding near optimum colour classifiers: genetic algorithm-assisted fuzzy colour contrast fusion using variable colour depth
This paper presents a complete Fuzzy-Genetic-based self-calibrating illumination intensity-invariant colour classification system. Previously, we have developed a novel fuzzy colour processing technique called Fuzzy Colour Contrast Fusion (FCCF) that selectively and adaptively corrects colours depic...
Gespeichert in:
Veröffentlicht in: | Memetic computing 2010, Vol.2 (3), p.219-236 |
---|---|
Hauptverfasser: | , |
Format: | Artikel |
Sprache: | eng |
Schlagworte: | |
Online-Zugang: | Volltext |
Tags: |
Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
|
Zusammenfassung: | This paper presents a complete Fuzzy-Genetic-based self-calibrating illumination intensity-invariant colour classification system. Previously, we have developed a novel fuzzy colour processing technique called Fuzzy Colour Contrast Fusion (FCCF) that selectively and adaptively corrects colours depicting target colour objects. FCCF has been proven to compensate for the effects of spatially varying illumination intensities in the scene, in various colour spaces. However, FCCF requires a huge set of parameters that is extremely tedious to calibrate by hand. To address these problems, we present a system that combines FCCF with a Heuristic-Assisted Genetic Algorithm (HAGA). FCCF-HAGA fully automates the fine-tuning of all colour descriptors, with significantly improved colour classification accuracy. Furthermore, we have reduced FCCF’s storage space requirements by processing colour channels selectively at varying colour depths. This is accomplished by combining a Variable Colour Depth (VCD) algorithm with FCCF that searches for the most effective colour depth for each colour channel. Our results show that for all cases, the FCCF-HAGA-VCD combination improves pie-slice colour classification. For six different target colours, under varying illuminations, the hybrid algorithm was able to yield 17.63% higher overall colour classification accuracy as compared to the pure fuzzy approach. Furthermore, it was able to reduce LUT storage space requirements by 78.06%, as compared to the full-colour depth LUT. |
---|---|
ISSN: | 1865-9284 1865-9292 |
DOI: | 10.1007/s12293-009-0025-8 |