Handling noise in single image defocus map estimation by using directional filters

TLDR: Experimental results on synthetic and real data demonstrate that the proposed new method for estimating a defocus map from a noisy image can estimate defocus maps better than the state-of-the-art approaches.

Abstract: State-of-the-art defocus map estimation methods are sensitive to image noise. Even a small amount of noise can degrade defocus map estimation dramatically. However, directly applying image denoising methods often changes edge profiles, thus leading to inaccurate defocus estimation. In this Letter, we propose a new method for estimating a defocus map from a noisy image. We observe that after using a directional low-pass filter to an input image, noise is greatly reduced while the edges orthogonal to the directional filter are well preserved. Based on this observation, we apply a series of directional filters at different orientations, and then estimate the blur amount of the edges, which are orthogonal to the direction of the filter in each filtered image. In order to obtain a full defocus map, we propagate the blur amount estimated at edges to the entire image by an edge-aware interpolation method. Experimental results on synthetic and real data demonstrate that our method can estimate defocus maps better than the state-of-the-art approaches.