Attention-Guided HDR Reconstruction for Enhancing Smart City Applications

Abstract

In the context of smart city development, video surveillance serves as a critical component for maintaining public safety and operational efficiency. However, traditional surveillance systems are often constrained by a limited dynamic range, leading to the loss of essential image details. To address this limitation, this paper introduces HDRFormer, an innovative framework designed to enhance high dynamic range (HDR) image quality in edge–cloud-based video surveillance systems. Leveraging advanced deep learning algorithms and Internet of Things (IoT) technology, HDRFormer employs a unique architecture comprising a feature extraction module (FEM) and a weighted attention module (WAM). The FEM leverages a transformer-based hierarchical structure to adeptly capture multi-scale image information. In addition, the guided filters are utilized to steer the network, thereby enhancing the structural integrity of the images. On the other hand, the WAM focuses on reconstructing saturated areas, improving the perceptual quality of the images, and rendering the reconstructed HDR images with naturalness and color saturation. Extensive experiments on multiple HDR image reconstruction datasets demonstrate HDRFormer’s substantial improvements, achieving up to a 2.7 dB increase in the peak signal-to-noise ratio (PSNR) and an enhancement of 0.09 in the structural similarity (SSIM) compared to existing methods. In addition, the framework exhibits outstanding performance in multi-scale structural similarity (MS-SSIM) and HDR visual difference predictor (HDR-VDP2.2). The proposed method not only outperforms the existing HDR reconstruction techniques but also offers better generalization capabilities, laying a robust foundation for future applications in smart cities.