| Summary: | 碩士 === 國立臺北科技大學 === 電子工程系研究所 === 103 === Haze removal techniques for visibility restoration in captured hazy images play a significant role in computer vision systems. However, to achieve real-time visibility restoration in these systems, effective hardware architecture is necessary for any haze removal technique. In this paper, we propose a hardware-oriented haze removal technique and its hardware architecture for visibility restoration. To achieve the goal of real-time application and low cost for hardware implementation, we propose several approximate techniques to replace high-complexity operations and avoid redundant computations. Furthermore, the proposed reconfigurable hardware architecture is used during hardware implementation to effectively avoid serious idle time and further reduce hardware cost. The experimental results indicate that the proposed hardware architecture not only provides more satisfactory restoration results than can other DCP-based techniques, but it also achieves an average frame rate of 32.15 fps at 1920×1080 image resolution.
|
|---|
