Saliency detection method for panoramic images based on GCN-ELM

In the field of saliency detection for panoramic images, traditional equirectangular and cube projection methods in panoramic image saliency detection often face issues like distortion and discontinuities, impacting detection accuracy. This study introduces an innovative image resampling technique and a GCN-ELM joint model. By evenly distributing spherical pixels onto a 2D plane, the method reduces pixel redundancy from equirectangular projection. Experimental results show that this approach significantly enhances saliency detection performance compared to existing methods.