Database retrieval for similar images using ICA and PCA bases

Naoto Katsumata, Yasuo Matsuyama

    Research output: Contribution to journalArticle

    36 Citations (Scopus)

    Abstract

    Similar-image retrieval systems are newly presented and examined. The systems use ICA bases (independent component analysis bases) or PCA bases (principal component analysis bases). These bases can contain source image's information, however, the indeterminacy of ordering and amplitude on the bases exists due to the PCA and ICA problem formulation per se. But, this paper successfully avoids this difficulty by using weighted inner products of similar bases. A set of opinion test is carried out on 18 systems according to the combination of {similarity measures (ICA, PCA, color histogram), color spaces (RGB, YIQ, HSV), filtering (with, without)}. The color histogram method is a traditional method. The opinion test shows that the presented method of {ICA, HSV, without filtering} is the best. Runners-up are {ICA, HSV or RGB or YIQ, with filtering}. The traditional method is judged to be much inferior. Thus, this paper's method is found quite effective to the similar-image retrieval from large databases.

    Original languageEnglish
    Pages (from-to)705-717
    Number of pages13
    JournalEngineering Applications of Artificial Intelligence
    Volume18
    Issue number6
    DOIs
    Publication statusPublished - 2005 Sep

      Fingerprint

    Keywords

    • ICA
    • Image bases
    • Independent component analysis
    • PCA
    • Principal component analysis
    • Similar image retrieval

    ASJC Scopus subject areas

    • Artificial Intelligence
    • Control and Systems Engineering

    Cite this