Salient object detection fusing global and local information based on nonsubsampled contourlet transform

Area of Science:

• Computer Vision
• Image Processing
• Signal Processing

Background:

• The nonsubsampled contourlet transform (NSCT) offers multiresolution, localization, directionality, and anisotropy.
• Directionality in NSCT is crucial for resolving intrinsic directional features in images.

Purpose of the Study:

• To develop a bottom-up salient object detection approach by integrating global and local image information.
• To leverage the NSCT for enhanced feature representation and saliency map generation.

Main Methods:

• Image decomposition using NSCT.
• Categorization and optimization of bandpass subband coefficients for improved representation.
• Generation of feature maps via inverse NSCT.
• Creation of global saliency maps based on feature likelihood.
• Calculation of local saliency maps using local self-information.
• Fusion of global and local saliency maps to produce the final saliency map.

Main Results:

• The proposed method demonstrates effectiveness in salient object detection.
• Experimental results on the MSRA 10K dataset show promising performance.
• The fusion of global and local saliency information yields superior results.

Conclusions:

• The developed NSCT-based approach provides an effective method for salient object detection.
• The fusion strategy significantly enhances the accuracy and robustness of saliency maps.
• The method shows potential for various computer vision applications requiring accurate object localization.