A parallel SIMD implementation of single image defogging

The increase of video cameras on phones, privately owned unmanned aerial vehicles, and now automobiles increase the demand for improved subjective quality in hazy and foggy weather. The presence of fog and haze reduce the local contrast of scenes which reduces the ability of humans to detect important objects. The most common method now to improve the contrast is to employ a single image defogging algorithm. Single image defogging has been a research topic of interest for the past seven years with newer advancements in complexity and color enhancements. There exists many methods with variations on the same theme of estimating a transmission map (thickness of fog) first and then removing the fog while respecting the true color of the scene. The methods that do exist, however, do not consider the algorithmic complexity constraints when implementing on a System on Chips (SoC). In our work, we demonstrate a single image defogging capability for live video on a SoC with parallel SIMD engine. We present the trade-offs and complications that need to be addressed for single image defogging methods for SIMD architecture hardware.