Laser & Optoelectronics Progress, Volume. 56, Issue 12, 121006(2019)

Saliency Detection Based on Background Continuity Priori Knowledge

Yikang Jin* and Fengqin Yu
Author Affiliations
  • College of Internet of Things, Jiangnan University, Wuxi, Jiangsu 214122, China
  • show less
    Figures & Tables(9)
    Typical local areas in image
    Shortest path among super-pixels
    Each stage for calculating foreground weight of each super-pixel of image based on background continuity. (a) Original image; (b) SLIC segmentation; (c) foreground weight before correction; (d) foreground weight after correction
    Partial saliency detection results of each method on ASD dataset
    PR curves and MAEs on three datasets. (a)(b) ASD; (c)(d) PASCAL-S; (e)(f) ECSSD
    Comparison of different methods on three datasets. (a) ASD; (b) PASCAL-S; (c) ECSSD
    Comparison of saliency results for image processed by three methods. (a) Original image; (b) based on global contrast; (c) based on local contrast; (d) based on background continuity
    • Table 1. Comparison of all index values obtained by three methods on ASD dataset

      View table

      Table 1. Comparison of all index values obtained by three methods on ASD dataset

      EvaluationindexLocalcontrastGlobalcontrastBackgroundcontinuity
      Precision0.86080.74580.8476
      Recall0.54050.69550.7411
      Fβ0.75720.73360.8204
    • Table 2. Comparison of running time for different methods

      View table

      Table 2. Comparison of running time for different methods

      MethodSF[13]GS[8]MR[9]SO[10]Ours
      Running time /s0.2650.2690.2700.3190.937
    Tools

    Get Citation

    Copy Citation Text

    Yikang Jin, Fengqin Yu. Saliency Detection Based on Background Continuity Priori Knowledge[J]. Laser & Optoelectronics Progress, 2019, 56(12): 121006

    Download Citation

    EndNote(RIS)BibTexPlain Text
    Save article for my favorites
    Paper Information

    Category: Image Processing

    Received: Dec. 4, 2018

    Accepted: Jan. 14, 2019

    Published Online: Jun. 13, 2019

    The Author Email: Jin Yikang (2362722976@qq.com)

    DOI:10.3788/LOP56.121006

    Topics