CNN 영상 회귀 기반의 산란계수 추정을 통한 연무제거

Abstract

The estimation of the scattering coefficient in depth image-based dehazing is of paramount importance. Since scattering coefficients are used to estimate the transmission image for dehazing, the optimal scattering coefficients for effective dehazing must be obtained depending on the level of haze and fog generation. In this study, we performed a CNN-based image regression to obtain the optimal scattering coefficients for each image with fog and haze. A three-channel image was used as the input data, and the learning was performed with approximately 2,000 labeled synthetic haze and fog datasets. Subsequently, the transmission image was estimated using the scattering coefficient obtained for the input image through the learned model, and the depth image was obtained through the LiDAR point cloud projection for performing the dehazing. This paper presents a qualitative and quantitative comparison of the results obtained using the proposed dehazing technique with those obtained using the existing dehazing algorithms.