Single Image Dehazing Algorithm Based on Modified Dark Channel Prior

Abstract

Single image dehazing algorithm based on Dark Channel Prior (DCP) is widely known. More and more image dehazing algorithms based on DCP have been proposed. However, we found that it is more effective to use DCP in the RAW images before the ISP pipeline. In addition, for the problem of DCP failure in the sky area, we propose an algorithm to segment the sky region and compensate the transmission. Extensive experimental results on both subjective and objective evaluation demonstrate that the performance of the modified DCP (MDCP) has been greatly improved, and it is competitive with the state-of-the-art methods.

Single Image Dehazing Based on Sky Area Segmentation and Image Fusion

“…We prove the performance of our method through quantitative and qualitative experiments, comparing it with stateof-the-art methods including DCP [8], Meng [19], Non-Local [10], MSCNN [13], DehazeNet [11], AODNet [12], He [20], Ehsan [21], Zhou [22], and D4 [14]. For quantitative evaluation, we adopt the widely used PSNR and SSIM [23] evaluation metrics.…”

“…( 2). DCP-based methods [22] have received a lot of attention due to its simplicity and effectiveness. However, DCP will fail in the sky area, which will generate noise and affect the visual effect.…”

“…However, DCP will fail in the sky area, which will generate noise and affect the visual effect. Different from the literature [22], we adopted a new segmentation method based on the characteristics of the sky region, proposed a new method for estimating atmospheric light and correcting transmission, and combined image enhancement methods to fuse multiple enhanced images to improve the quality of human visual perception.…”

Ancient character recognition with deep learning techniques

“…Deep neural networks effectively improve the prediction accuracy, while the extremely high requirements for data with effective labeling make its training costs rise. The real situation is variable, and there may be noise interference such as light, angle, and distortion [7,8]. As the number of parameters increases, the computing time also increases, which means more computing power is needed.…”