Roy Orfaig holds a B.Sc. in Bio Medical Engineering (2005) and a M.Sc. in Electrical Engineering (2010) from Ben-Gurion University. He joined Applied Materials in 2005 and during six years filled a number of key roles in the development of various detection algorithms for mask and wafer inspection. In 2011 he joined an IAI image processing department and since then has worked as an algorithms developer and project manager in various fields in computer vision, image and video processing, machine learning and signal processing. Since 2013 he served as supervisor of final projects in Electrical Engineering department as part of cooperation between IAI and Tel-Aviv University.
Advanced Algorithm of Haze Removal Based on Dark Channel Prior
The dark channel prior method of haze removal as presented by He et al (2009). provides a novel solution to the problem of haze and mist in a single outdoors image. However, it is limited in its scope and inefficient in its implementation for a stationary model. Haze is spatially uniform and consistent in the RGB color space, against other phenomena like smoke which is not is not stationary and spatially variable.
We propose to expand the dark channel prior method in order to improve the algorithm performance and also provide a solution to a range of atmospheric phenomena (e.g. smoke), as well as optimize its implementation in order to make it more feasible as a video processing algorithm. Processing a video by executing the algorithm frame-by-frame creates a strong flickering effect due to the varying results of the airlight (atmospheric value) estimations between frames. We overcome this problem by developing an adaptive airlight estimator which keeps a smooth frame progress.
This research was part of cooperation between IAI and Tel Aviv University as a student project in the Electrical Engineering department. The project won the Prof. Udi Weinstein reward as the best final project in Signal processing for 2013/4.