FAST ALGORITHMS FOR MULTI-WINDOWED DARK CHANNEL PRIORS WITH LESS COMPARISON AND TIME CYCLE REQUIREMENT
Journal
IET Conference Proceedings
Journal Volume
2023
Journal Issue
35
ISBN
978-1-83953-940-4
Date Issued
2023-01-01
Author(s)
Abstract
The dark channel prior (DCP) is important in image dehazing. Since for a natural image the dark channel value is always near to zero, the DCP is helpful for estimating the transmission and retrieve the original image. However, since the DCP is nonlinear and space variant, it requires a huge amount of comparisons, especially in the multi-window case. In this paper, an advanced DCP implementation algorithm is proposed. We apply the fact that, after taking the minimum of 3 color channels, the DCP is the closing operation with a convex kernel. Thus, we can use an iterative algorithm to compare the rows (columns) with the displacement of 2q. With the proposed algorithm, the number of comparisons is only 2+2ceil(log2L) for each pixel where LxL is the patch size and the required memory is only 3MN where 3xMxN is the size of the color image.
Subjects
dark channel prior | dehazing | fast algorithm | image processing | multiple windows
Type
conference paper