דומיניק ברטושט (אונ' אלנגן, גרמניה)
חדר 601, בניין טאוב למדעי המחשב
The K-SVD algorithm is a powerful tool for image denoising but
computationally quite intensive. In order to accelerate it we
implement the most time consuming parts like sparse coding and
assembling the final denoised image vectorized and in parallel on the
Cell processor. The Cell processor is a heterogeneous multicore system
consisting of a general purpose Power-PC processor unit (PPU) and
several fast co-processors, the synergistic processor elements (SPEs).
We present runtime results comparing a standard CPU and the Cell
implementation and apply the denoising algorithm to CT images. Here,
one goal is to estimate the non-gaussian, non-stationary noise in the
images and include this information in the K-SVD algorithm.