A novel multiresolution spatiotemporal saliency detection model and its applications in image and video compression

Area of Science:

• Computer Vision
• Image Processing
• Computational Neuroscience

Background:

• Object detection relies on identifying salient areas, areas the human eye focuses on.
• Existing saliency detection models like STB and NVT are computationally expensive and parameter-dependent.
• Previous methods, including spectral residual (SR), often fail to achieve real-time performance.

Purpose of the Study:

• To propose a novel, computationally efficient, and real-time saliency detection model.
• To enhance saliency detection for both static images and dynamic videos.
• To improve image and video compression efficiency using saliency information.

Main Methods:

• Developed a quaternion representation incorporating intensity, color, and motion features.
• Introduced the Phase Spectrum of Quaternion Fourier Transform (PQFT) model based on Fourier transform phase spectrum principles.
• Proposed the Hierarchical Selectivity (HS) framework and Multiresolution Wavelet Domain Foveation (MWDF) for image/video compression.

Main Results:

• The PQFT model demonstrates superior effectiveness in saliency detection compared to existing state-of-the-art models.
• PQFT accurately predicts human eye fixations across various resolutions.
• The HS-MWDF model achieves higher compression rates in image and video compression.

Conclusions:

• The PQFT model offers a low-computational cost solution for real-time spatiotemporal saliency detection.
• The proposed model significantly improves saliency detection accuracy and eye fixation prediction.
• The integration of PQFT with HS-MWDF enhances compression efficiency for multimedia content.