ICA Research Website

Single Image Dehazing using a Multilayer Perceptron



This paper presents a novel algorithm to improve images with hazing effects. Usually the dehazing methods based on the dark channel prior, make use of two different stages to compute the transmission map of the input image. The stages are the transmission map estimation and a transmission map refinement. However, the main disadvantage of these strategies is the trade-off between accurate restoration and computational time. The proposed method uses a Multilayer Perceptron to compute the transmission map directly from the minimum channel, and a contrast stretching technique to improve the dynamic range of the restored image. The Multilayer Perceptron is trained in terms of Mean Squared Error (MSE) using a training set of 80 images. To evaluate the restoration quality, the metrics of Mean Absolute Error (MAE), and the Structural Similarity (SSIM) index are used. The experimental results have proven that the proposed method achieves superior performance in terms of restoration quality (MAE = 27.28, SSIM index = 0.84) compared with seven state-of-the-art dehazing methods. In addition, based on the average computational time achieved by the proposed method (0.52 seconds using a test set of 40 images), it can be highly suitable for real-time applications.

Authors:

Sebastian Salazar-Colores, Ivan Cruz-Aceves*, and Juan Ramos-Arreguin

Research paper published by Journal of Electronic Imaging, SPIE (March,2018)

Journal of Electronic Imaging, 27(4), 043022 (2018), https://doi.org/10.1117/1.JEI.27.4.043022

Experimental results

Time performance comparison

Fig. 1 Performance analysis in terms of computational time.

He et al. [1], Tarel et al. [2], Pang et al. [3], Ren et al. [4], Berman et al. [5], Zhu et al. [6], Gibson et al. [7], Proposed method

Visual analysis

Image Cones

Original

He et al. [1]

Tarel et al. [2]

Pang et al. [3]

Ren et al. [4]

Berman et al. [5]

Zhu et al. [6]

Gibson et al. [7]

Proposed method

Image Forest

Original

He et al. [1]

Tarel et al. [2]

Pang et al. [3]

Ren et al. [4]

Berman et al. [5]

Zhu et al. [6]

Gibson et al. [7]

Proposed method

Image Ny_17

Original

He et al. [1]

Tarel et al. [2]

Pang et al. [3]

Ren et al. [4]

Berman et al.[5]

Zhu et al. [6]

Gibson et al. [7]

Proposed method

Downloads

Matlab source code

References

[1] K. He, J. Sun, and X. Tang, “Single Image Haze Removal Using Dark Channel Prior,” IEEE Transactions on Pattern Analysis and Machine Intelligence 33(12), 2341–2353 (2010).
[2] P. Tarel and N. Hautiere, “Fast visibility restoration from a single color or gray level image,” in IEEE International Conference on Computer Vision (ICCV), 1, 2201–2208, (Kyoto, Japan) (2009).
[3] J. Pang, A. Oscar, and G. Zheng, “Improved single image dehazing using guided filter,” in Proceedings of the APSIPA Annual Summit and Conference (APSIPA ASC), 1, 1–4, (Xi’an, China) (2011).
[4] W. Ren, S. Liu, H. Zhang, et al., "Single Image Dehazing via Multi-scale Convolutional Neural Networks", 154–169. Springer International Publishing, Cham (2016).
[5] Berman, T. Treibitz, and S. Avidan, “Non-Local Image Dehazing,” in The IEEE Confer ence on Computer Vision and Pattern Recognition (CVPR), (2016).
[6] Zhu, J. Mai, and L. Shao, “A Fast Single Image Haze Removal Algorithm Using Color Attenuation Prior,” IEEE Attenuation Prior,”, IEEE Transactions on Image Processing 24, 3522–3533 (2015).
[7] B. Gibson, D. T. V ˜o, and T. Q. Nguyen, “An investigation of dehazing effects on image and video coding,” IEEE transactions on image processing : a publication of the IEEE Signal Processing Society 21(2), 662–73 (2012).