Fast Execution Schemes for Dark-Channel-Prior-Based Outdoor Video Dehazing

Abstract

This paper studies the dark-channel-prior (DCP)-based dehazing from the implementation perspectives. Several schemes are proposed, in order to realize the fast execution of the DCP-based method targeting the outdoor video dehazing. The airlight estimation is performed based on distributed sorting with scaled luminance and executed selectively by considering the coherence between successive images in a video stream, which results in a significant reduction of complexity. In addition, the complexity involved in the transmission estimation is reduced by an assumption regarding the chrominance of the airlight under the outdoor environment. The performance of the DCP-based dehazing method with the proposed schemes is evaluated under a prototype implementation in terms of execution time, as well as dehazing quality. In the DCP-based dehazing method with the proposed schemes, the overall execution time is reduced by up to 49% while the dehazing quality is maintained to that of the original method. The demonstration video and the experimental results are provided publicly.

Keywords

• Dehazing
• fast execution
• dark channel prior
• embedded systems
• haze removal
• image enhancement