A Retinex-Based Local Tone Mapping Algorithm Using L0 Smoothing Filter - Advances in Image and Graphics Technologies - page 46

The parameter ʻ controls the smoothness of image, and a small ʻ produces better
dynamic range compression but reduces the global contrast. The parameter ʲ controls
the overall brightness of the image, and a large ʲ makes the result image brighter. The
parameter ʵ is the nonlinearity offset of logarithm function, and a small ʵ can
effectively enhance the contrast of dark areas. The parameter s controls the color
saturation of the image.
The comparison of our algorithm with other tone mapping operators is presented in
Fig. 2. In order to make the comparison fairly, all the methods are tested with default
parameters. Since there is no computational model to evaluate tone mapping
algorithms objectively, we evaluate the results in a subjective way. Obviously, our
algorithm shows a better performance in terms of naturalness and local contrast than
other three algorithms. The middle row and the bottom row in Fig. 2 are respectively
the enlarged images of bright and dark areas, and the result of our algorithm shows
the clearest details of the four algorithms. In a word, our algorithm shows good color
rendition while maintaining good contrasts elsewhere, and at the same time reduces
halo artifacts.

5

Conclusion

In order to compress the dynamic range of the HDR images for display, a local tone
mapping algorithm is proposed in this paper. In the global adaptation, we adopt
HALEQ to enhance the global contrast. The local adaptation is based on retinex
theory, and we use L 0 smoothing filter as surround function instead of Gaussian
function to reduce the halo artifacts. The global and local tone mapping is only
applied to luminance. Finally, the tone mapped image is obtained from the processed
luminance and the original RGB image through color correction. The experimental
results demonstrate that the proposed method effectively compress the dynamic range
of image while with good color rendition and detail preservation.
However, since the proposed method uses L 0 smoothing filter in local tone
mapping, it will be much slower compared with the global tone mapping algorithms.
Future work is to optimize the computational model and also try to realize the parallel
algorithm on GPU.

References

1. Larson, G.W., Rushmeier, H., Piatko, C.: A Visibility Matching Tone Reproduction
Operator for High Dynamic Range Scenes. IEEE Transactions on Visualization and
Computer Graphics 3, 291-306 (1997)
2. Land, E., McCann, J.: Lightness and Retinex Theory. J. Opt. Soc. Amer. 61(1), 1-11
(1971)
3. Jobson, D.J., Rahman, Z., Woodell, G.A.: Properties and Performance of a
Center/Surround Retinex. IEEE Trans. Image Processing 6(3), 451-462 (1997)
4. Xu, L., Lu, C., Xu, Y., Jia, J.: Image Smoothing via L 0 Gradient Minimization. ACM
Trans. Graph 30, 6 (2011)