Efficient image denoising algorithm with block-matching and 3D filtering

  • Algorithm. We develop a two-step algorithm.

    • The hard-thresholding approach delivers an initial estimate.

    • The Wiener filtering approach to produces the final estimate.


  • Transform choice. In our experiments, we used 2D DFT and 3D DFT for $\QTR{cal}{T}_{2D}$ and $\QTR{cal}{T}_{3D}$, respectively.


  • Complexity reduction. Efficient trade-off between denoising performance and computational complexity. Constarints:

    • Restrict the maximum number of matched blocks.

    • Perform block-matching within a local neighborhood of fixed size, rather than doing it in the whole image.

    • Use a step greater than unity to slide to every next reference block.


  • Running time. The execution time of the algorithm that we used for the reported results is about 8 seconds for an image of size 256x256, on a 3 GHz Pentium workstation.