As the classic image segmentation technique, Otsu adaptive thresholding method is widely employed in image processing and computer vision. Two-dimensional Otsu thresholding algorithm was regarded as an effective improvement of the original Otsu method, especially under low SNR condition. However, the tremendous computation cost limited the application of two-dimensional Otsu algorithm in real-time situation. This paper addresses this problem and proposes a novel fast algorithm. We utilized integral image to simplify the redundant calculation for searching optimal threshold. Instead of repeating accumulations, the sum of rectangle area can be calculated by several addition operations. Both the theoretical analysis and segmentation experiment proved that the proposed approach can significantly reduce the computation cost and still obtain the identical optimal threshold acquired by original two-dimensional Otsu algorithm.


    Zugriff

    Zugriff prüfen

    Verfügbarkeit in meiner Bibliothek prüfen

    Bestellung bei Subito €


    Exportieren, teilen und zitieren



    Titel :

    Integral Image Based Fast Algorithm for Two-Dimensional Otsu Thresholding


    Beteiligte:
    Lang, Xianpeng (Autor:in) / Zhu, Feng (Autor:in) / Hao, Yingming (Autor:in) / Ou, Jinjun (Autor:in)


    Erscheinungsdatum :

    01.05.2008


    Format / Umfang :

    687596 byte




    Medientyp :

    Aufsatz (Konferenz)


    Format :

    Elektronische Ressource


    Sprache :

    Englisch





    Research on Image Edge Detection Method Based on Improved Robinson Operator and Otsu Algorithm

    Ren, Hao / Sun, Jie / Yuan, Wei | British Library Conference Proceedings | 2020


    Multilevel thresholding by fast PNN-based algorithm [5014-14]

    Virmajoki, O. / Franti, P. / Society for Imaging Science and Technology | British Library Conference Proceedings | 2003


    Fast algorithm for 2D entropic thresholding of image segmentation [4550-64]

    Su, J. / Tian, J. / Liu, J. et al. | British Library Conference Proceedings | 2001