A Modified DCP Based Dehazing Algorithm

Abstract

In the research field of haze removal, a popular model-based scheme is based on the dark channel prior (DCP) because of its simplicity. However, it is well-known that the DCP scheme in [1-2] suffers from four problems: (i) artifacts happened in sky regions; (ii) halos occurred in large depth discontinuities; (iii) color distortions found in recovered images; (iv) high computational cost required. To relieve the problems in the DCP scheme, this paper presents a modified DCP scheme (MDCP). The proposed MDCP scheme follows similar implementation steps in [1-2] except the estimations for the atmospheric light and the transmission map are modified. In atmospheric light estimation, a pixel-based dark channel is employed with an adaptive scaling factor. In the transmission map estimation, an adaptive scaling factor is to replace the fixed scaling factor to find the initial transmission map. Moreover, the pixel-based dark channel is used as the guide image in the guide image filtering whereas the DCP scheme in [2] employs the input image. By the above modifications, the proposed MDCP is verified by examples and compared with the DCP scheme by subjective and objective assessments. The simulation results indicate that better subjective and objective assessments are for the proposed MDCP scheme while the aforementioned problems in the DCP scheme are relieved. As for the efficiency, the proposed MDCP scheme, on average, is 5.12 times faster than the DCP scheme in the given examples.