Vectorization and optimization of fog removal algorithm

Some of the image processing algorithms are very costly in terms of operations and time. To use these algorithms in real-time environment, optimization and vectorization are necessary. In this paper, approaches are proposed to optimize, vectorize and how to fit the algorithm in low memory space. Here, optimized anisotropic diffusion based fog removal algorithm is proposed. Fog removal algorithm removes the fog from image and produces an image having better visibility. This algorithm has many phases like anisotropic diffusion, histogram stretching and smoothing. Anisotropic diffusion is an iterative process that takes nearly 70% of time complexity of the whole algorithm. Here, optimization and vectorization of the anisotropic diffusion is proposed for better performance. However, optimization techniques cost some accuracy but that can be neglected for significant improvement in performance. For memory constraint environment, a method is proposed to process the entire block of image and maintains the integrity of operations. Results confirm that with our optimization and vectorization approaches, performance is increased up to 90 fps (approximately) for VGA image on one of the image processing DSP simulator. Even if, system doesn't have vector operations, the proposed optimization techniques can be used to achieve better performance (2x faster).