Massively Parallel Exact Histogram Equalization
Event Type
ACM Student Research Competition: Graduate Poster
ACM Student Research Competition: Undergraduate Poster
Posters
Student Program
TP
XO
TimeThursday, 19 November 20208:30am - 5pm EDT
LocationPoster Module
DescriptionHistogram specification is performed by transforming an image so that the image's histogram matches a target histogram. Exact methods of histogram specification result in less loss of information but are orders of magnitudes slower than the classical methods. Applications such as real-time medical imaging that require histogram equalization currently must use the classical method. This project adapts exact histogram equalization methods to run on GPUs to greatly increase their speed and make the use of exact histogram equalization methods viable for real-time imaging applications along with increasing the speed of large-scale machine learning applications. Some of the techniques are applicable to any exact histogram specification method, but two of the most popular methods (local means and varational) are particularly optimized. The optimized algorithms running on a GPU provide super-real-time speeds on very large images while maintaining the quality of the results. Comparisons of speedups will be shown.
