Detection of Images for Video Based Spectroscopy by Using the Combination of Saliency Methods

Depth perception affects 2D image brightness, color, texture, and motion. In addition to stereoscopic vision, depth range, displaying size, 3D visualization, naturalness, and visual comfort can reconstruct 3D depth information. These features do not affect 2D visual attention. Replicating human visual attention in 3D situations requires saliency prediction methods designed for 3D environments. Stereoscopic videos' visual contrasts and object movements make saliency detection difficult. Depth map data is essential for saliency detection. For practical use, stereoscopic data is better than 2D. As visual scenes get more complex, 2D data alone cannot identify significant items. Depth maps help distinguish related items in complex situations. We use spatiotemporal correlation and depth information refinement to overcome these problems in this saliency detection investigation. Comparative analytical analysis of a baseline spatiotemporal depth correction system for multimodal saliency detection frameworks supports the basic premise. We chose two more saliency detection models: a cutting-edge multimodal filter transformer-based approach and a two-stream hierarchical model. We then compared these models to the baseline. We found that the spatiotemporal saliency model is effective due to its higher mean absolute error (MAE) values. Saliency models show that this method reduces visual region estimation and object detection mistakes. The spatiotemporal depth correction strategy is proven necessary and successful by this method.