An Efficient FPGA Implementation of Optimized Anisotropic Diffusion Filtering of Images

Figure 17

Comparison of various edge-preserving filters implemented using hardware on the natural grayscale image Einstein of size 150 × 150 measured for similar number of iterations. (a) Original image. (b) Direct implementation of the anisotropic diffusion filter (software implementation). (c) Output of the bilateral filter realized using FPGA. (d) Output of nonlocal means filter. (e) Trilateral filter output. (f) Output of our implementation of optimized anisotropic diffusion filtering using a novel hardware design. Note that the fine boundary transitions in the moustache area (see zoomed insets) are clearly visible in our implementation in (f) unlike others which is clear from the visual experience. Similarly, the zoomed portions of the left eye show the clear distinctions of the lower and upper lid (also holds the contrast information); moreover, the magnified area of the neck portion also shows a sharp transition. All the comparisons should be done keeping the original image in (a) in mind as the reference image. The PSNR difference is as shown in Table 2.

(a)

(b)

(c)

(d)

(e)

(f)