심도정보의 Entropy를 이용한 개선된 DCP안개 제거 방법

Abstract

It is important to haze removal in image processing because it makes it difficult to analyze color information and edge information in marine and aeronautical fields, which are sensitive to meteorological conditions. The DCP(Dark Channel Prior), which estimates the haze using the minimum values of R, G, B information, is the most widely used algorithm to remove haze from the current image information. The DCP algorithm is a method for estimating the amount of haze by using the minimum value of R, G, B information on a local area selected stepwise from a given fog image, and estimating the transmission map to remove the haze. At this time, the haze is estimated from the edge of the boundary to the local area, so that the block artifact inevitably occurs. Therefore, the image analysis performance is not high near the edge.

This paper proposes a haze removal method using an improved transmission map to reduce the block artifact occurred during DCP process which is a representative algorithm for haze removal. The proposed method estimates depth information and edge information in a dark channel using entropy, which are stochastic properties, and predicts the part where block artifact occurs. Using the adaptive window according to the entropy value in the predicted part, new transmission map is obtained, which can reduce the block artifact in the edge of the boundary containing the depth information.

In the conclusion, we can obtain improved fog removal image than transmission map the existing DCP algorithm by using the new transmission map.