Search papers, labs, and topics across Lattice.
This paper introduces a depth-based method for image and video orientation estimation, partitioning images into quadrants and analyzing depth distributions to infer orientation. Depth Gradient Consistency (DGC) and Horizontal Symmetry Analysis (HSA) are incorporated to refine orientation estimates and improve perceptual alignment. Experiments demonstrate the proposed approach outperforms existing methods in diverse scenarios, showcasing robustness and accuracy.
Forget feature-based orientation estimation – depth cues offer a surprisingly robust and accurate alternative for aligning images and videos.
This paper introduces a novel approach for image and video orientation estimation by leveraging depth distribution in natural images. The proposed method estimates the orientation based on the depth distribution across different quadrants of the image, providing a robust framework for orientation estimation suited for applications such as virtual reality (VR), augmented reality (AR), autonomous navigation, and interactive surveillance systems. To further enhance fine-scale perceptual alignment, we incorporate depth gradient consistency (DGC) and horizontal symmetry analysis (HSA), enabling precise orientation correction. This hybrid strategy effectively exploits depth cues to support spatial coherence and perceptual stability in immersive visual content. Qualitative and quantitative evaluations demonstrate the robustness and accuracy of the proposed approach, outperforming existing techniques across diverse scenarios.
