Tecnalia has presented new results on image restoration in a paper entitled “A Variational Framework for Image Dehazing” in this year European Conference on Computer Vision. This work is fruit of joint collaboration with Pompeu Fabra University at Barcelona and the Basque Country University.
Image dehazing consists on the restoration of images captured under adverse weather conditions, such as haze or fog. These images typically exhibit low contrast and faded colors, which may severely limit the visibility within the scene. Unveiling the image structure under the haze layer and recovering vivid colors out of a single image remains a challenging task, since the degradation is depth-dependent and conventional methods are unable to handle this problem.
The developed work extends a well-known perception-inspired variational framework for the task of single image dehazing. The previous framework consisted on minimizing the following energy:
where I is the degraded image. The result of this minimization outputs an enhanced version of the initial image that complies with several assumptions of the Human Visual System. However, this energy is not well adapted to the problem of image dehazing, since it implements the Gray World assumption.
We propose to replace the value used by this framework for the grey-world hypothesis by an estimation of the mean of the clean image. This allows us to devise a variational method that requires no estimate of the depth structure of the scene, performing a spatially-variant contrast enhancement that effectively removes haze from far away regions.
To estimate a suitable value for the mean of the clean image, we compute:
where A is an estimate of the color of the haze, that is easily obtained from the degraded image, and L is the luminance of the degraded scene. Here also, j denotes channelwise computations.
Experimental results show that our method competes well with other state-of-the-art methods in typical benchmark images, while outperforming current image dehazing methods in more challenging scenarios.
The main advantage of our novel method, when compared to other approaches, is that we are more robust to changes in the global illuminant of the scene. This is obtained thanks to the non-dependence of our algorithm from a previous computation of a depth map of the scene, which is usually based on too simplistic physical models.
Tecnalia is currently working on implementing these algorithms on dedicated DSPs for different industrial applications.