Adaptive Seam Optimization in Image Stitching via Saliency-Based Weighting and Smoothness Constraints

Image stitching plays a pivotal role in various computer vision applications. This paper proposes an adaptive seam optimization method to address issues such as misalignment, ghosting, and the loss of significant details in highly salient regions due to poor seam placement. Our approach integrates saliency-based weighting with smoothness constraints, dynamically adjusting the seam position to avoid cutting through visually important regions. Additionally, smoothness constraints evaluate color, texture, and edge differences between adjacent pixels, ensuring smooth transitions across seams. Experimental results demonstrate the effectiveness of the proposed method, showing a significant reduction in ghosting and misalignment compared to classical and state-of-the-art stitching techniques. Our method excels particularly in highly salient regions and complex scenes, producing high-quality seamless panoramic images.