Randomized Redundant DCT:
Efficient Denoising by Using Random Subsampling of DCT Patches

Shu Fujita, Norishige Fukushima , Makoto Kimura, Yutaka Ishibashi
RR-DCT

In this paper, we propose an acceleration method for image denoising with redundant discrete cosine transform (R-DCT). Image denoising is essential for image processing, and its efficiency is important for graphics applications. R-DCT with hard-thresholding can perform denoising while keeping detail textures. Moreover, the method is computationally efficient compared with state-of-the-art denoising methods, such as BM3D. The computational cost, however, is still insufficient for real-time processing; hence, we accelerate the method by using randomized subsampling of DCT patches. Experimental results show that our method can accelerate the processing, while the degradation of denoising performance is a little.

The link is our code on Github. The implimentation is written in C++ with SIMD vectorization (SSE) and multi-threading. The code requireds OpenCV.

  1. S. Fujita, N. Fukushima, M. Kimura, and Y. Ishibashi, "Randomized Redundant DCT: Efficient Denoising by Using Random Subsampling of DCT Patches," SIGGRAPH Asia 2015 Technical Briefs, Nov. 2015.
@inproceedings{fujita2015siggraphasia_tb,
    author  = {S. Fujita and N. Fukushima and M. Kimura and Y. Ishibashi},
    title   = {Randomized Redundant DCT: Efficient Denoising by Using Random Subsampling of DCT Patches},
    booktitle = {SIGGRAPH Asia 2015 Technical Briefs},
    year    = {2015},
}