high performance computing on graphics processing units: hgpu.org

Digital Image Processing is an evergreen area of research in the signal processing domain. Denoising of digital images is one of the most fundamental operations that is performed in the pre-processing stage of almost all image processing operations. This important feature makes denoising as one of the lucrative research areas within the broad area of Digital Image Processing. With the advancement of upcoming hardware technologies, the demand of scientific computing can be better addressed by the image processing researchers. The clock frequency of a typical CPU has hit a ceiling at around 3.5 GHz, and the trend of the CPU market is now shifted towards providing more numbers of processing cores rather than higher clock frequencies. Under such scenario, it becomes necessary for a programmer to develop parallelized software algorithms, to fully exploit the multicore features of the available hardware The latest trend in advanced computation platforms is the General Purpose Graphical Processing Unit (GPGPU) computing, where the conventional use of GPU’s for gaming purpose alone is extended greatly, and exploited to solve general engineering problems of high dimensionality. As long as any problem has the ability to be parallelized, it can be optimized for implementation on the GPU. Therefore, the use of GPU for implementation of the bilateral filtering [1] algorithm is proposed, since the latter is an inherently parallelizable algorithm. As a practical application, the algorithm is proposed to be used for analyzing its denoising of digital images photographed in low light conditions. Preliminary results for Bilateral Filtering implementation using Matlab strongly favor the further development of this approach for the said application [3].