Robust color image segmentation by Karhunen-Loeve transform based Otsu multi-thresholding and K-Means Clustering

Chenxue Wang, Junzo Watada

    Research output: Chapter in Book/Report/Conference proceedingConference contribution

    7 Citations (Scopus)

    Abstract

    In this paper, a novel fast approach is proposed to achieve image segmentation in color image. This method helps to refine the foreground regions and achieves the goal of robust color image segmentation throw the following four steps. First, modified Karhunen-Loeve transform is performed to reduce the redundant component, thus selecting the most important part of the color images. Second, a multi-threshold Otsu method is carried out to select the best thresholds from image histogram. Thereby, the conventional Otsu method has been extended from gray level to color level. Third, improved Sobel edge detection is added to enhance the weight of edge detail of the foreground image. Finally, a K-Means Clustering is used to merge the over-segmented regions. Experimental results prove that this method has a good performance even when the color image has a complicated structure in the background.

    Original languageEnglish
    Title of host publicationProceedings - 2011 5th International Conference on Genetic and Evolutionary Computing, ICGEC 2011
    Pages377-380
    Number of pages4
    DOIs
    Publication statusPublished - 2011
    Event5th International Conference on Genetic and Evolutionary Computing, ICGEC2011 - Xiamen
    Duration: 2011 Aug 292011 Sep 1

    Other

    Other5th International Conference on Genetic and Evolutionary Computing, ICGEC2011
    CityXiamen
    Period11/8/2911/9/1

    Keywords

    • Background subtraction
    • Image segmentation
    • K-Means Clustering
    • Karhunen-Loeve transform
    • Otsu method

    ASJC Scopus subject areas

    • Computational Theory and Mathematics
    • Computer Science Applications

    Fingerprint Dive into the research topics of 'Robust color image segmentation by Karhunen-Loeve transform based Otsu multi-thresholding and K-Means Clustering'. Together they form a unique fingerprint.

    Cite this