Abstract:
Fields of image processing and computer vision includes many computationally
expensive algorithms and they take too much time in their execution due to their serial
implementation. Speedup in computations is either achieved by increasing the clock speed
or by increasing the number of cores. Since clock speed has reached its limit, many core
strategy is the only way to achieve the speedup. Now a days GPUs are widely used for
highly computational tasks due to their many core throughput oriented design.
Underwater water video enhancement is impossible to run in real-time on CPU due
to expensive computational requirements. We accelerated the process of underwater video
enhancement by using CUDA programming model (NVIDIAs GPU programming model).
We made a parallel implementation of enhancement steps like generation of Inputs of
fusion process, calculation of weight maps of fusion process and the final blending process.
We made the process of underwater video enhancement in Real-time by getting more than
10x speed up in enhancement process. Now it is possible to enhance the underwater videos
in real time by using our GPU implementation. We used low end graphics processing unit
with just 96 cores, more speedup can be gained by using latest graphics card.
