Title : Fusion Using Laplacian Pyramid Transform of Light & Dark Channel Priors for Image De-Hazing
Authors : Karuna Khatter1, Bhupender Sharma2, Ravi Malik3
This paper proposes an effective and efficient method for haze removal from a single input image. The method employs both techniques i.e. dark and light channel priors. DCP (Dark channel prior), a scheme for dehazing images represents a statistical property for dehazed single images i.e. most patches for an image contains dark pixels for atleast single color channel. The two main limitations for this method are: 1. For bright image patches, there is over-exposure for atmospheric light, 2. Soft mapping technique applied for dark channel prior to compute factor t (transmission map) costs high. Hence another dehazing algorithm employing both techniques dark and light channel is proposed in this paper where light channel represents a statistics for hazy outdoor images. Furthermore, the guided and bilateral filter are employed for the dark and light channel image refinement. This dehazing algorithm alleviates the above mentioned limitations with Dark channel prior. This paper shows several examples for the comparison for proposed work with the existing Dark channel prior. Further the filters added refine the image to a greater extent. The results comparison prove that this algorithm is 25 times faster than the Dark channel prior. Also the visual quality is much better with the employment of proposed bilateral filter. Therefore the haze effects can be reduced to a great extent and can be utilized for many applications such as intelligent transportation system, aircrafts system, video surveillance, remote sensing etc.