Graph-Based Visual Saliency (GBVS) Algorithm Implementation

This documentation provides MATLAB implementations of Graph-Based Visual Saliency (GBVS), a graph-based approach for computing image saliency maps. The code package includes three main algorithm variants: the complete GBVS implementation [gbvs] featuring multi-scale feature extraction and Markov chain-based activation maps, a computationally efficient simplified version [simpsal] with optimized graph construction, and an Image Signature-based algorithm [signatureSal] utilizing spectral analysis for saliency detection. For installation, copy the decompressed files to MATLAB's toolbox directory and add the path using MATLAB's "addpath" function. The modular architecture allows straightforward customization for specific applications - users can extend image preprocessing pipelines, optimize algorithm parameters through configuration files, or integrate additional feature extraction methods. Each algorithm returns saliency maps as 2D matrices suitable for further computer vision tasks like object detection and image segmentation. The implementation follows MATLAB best practices with vectorized operations for performance efficiency. Key functions include graph construction using Gaussian dissimilarity metrics, random walker algorithms for saliency propagation, and multi-scale pyramid processing. Researchers can modify kernel parameters, adjust graph connectivity thresholds, or implement custom normalization schemes to adapt the algorithms for specific datasets or performance requirements. This toolbox provides a flexible foundation for image saliency research with extensible architecture supporting both academic and industrial applications.