Deblurring Face Images with Exemplars

Abstract

Abstract. The human face is one of the most interesting subjects involved in numerous applications. Significant progress has been made towards the image deblurring problem, however, existing generic deblurring methods are not able to achieve satisfying results on blurry face images. The success of the state-of-the-art image deblurring methods stems mainly from implicit or explicit restoration of salient edges for kernel estimation. When there is not much texture in the blurry image (e.g., face images), existing methods are less effective as only few edges can be used for kernel estimation. Moreover, recent methods are usually jeopardized by selecting ambiguous edges, which are imaged from the same edge of the object after blur, for kernel estimation due to local edge selection strategies. In this paper, we address these problems of deblurring face images by exploiting facial structures. We propose a maximum a posteriori (MAP) deblurring algorithm based on an exemplar dataset, without using the coarse-to-fine strategy or ad-hoc edge selections. Extensive evaluations against state-of-the-art methods demonstrate the effectiveness of the proposed algorithm for deblurring face images. We also show the extendability of our method to other specific deblurring tasks.

Exploiting Semantics for Face Image Deblurring

“…Table 9 presents the average PSNR and SSIM for different sizes of blur kernels, and Table 10 shows the average and the worst PSNR/SSIM on the entire datasets for each method. We note that the optimization-based methods (Shan et al, 2008;Cho and Lee, 2009;Krishnan et al, 2011;Xu et al, 2013;Zhong et al, 2013;Pan et al, 2014Pan et al, , 2017a) may generate severe visual artifacts when the blur kernel is not estimated well and achieve significant lower PSNR/SSIM values. The proposed method performs favorably against existing deblurring approaches and our preliminary method on both datasets.…”

“…Numerous approaches exploit domain specific visual information, such as designing L 0 intensity (Pan et al, 2017b) priors for text images or detecting light streaks (Hu et al, 2014a) for extremely low-light images. As face images contain fewer textures and edges for estimating blur kernels, Pan et al (2014) search for similar face exemplars from an external dataset and extract the contour as reference edges. However, reference images may not always exist for a specific input due to diversity of real-world face images.…”

Privacy preserving optics for miniature vision sensors

“…Face De-blurring. Despite significant advances in image and video de-blurring [54,72,51,50,24,3,14], de-blurring heavily blurred images is still an open problem. In this paper, some designs that use optical defocus for privacy may be susceptible to reverse engineering.…”

Discriminative learning of iteration-wise priors for blind deconvolution

“…As suggested in [4,32,IS,21], we formulate the pro posed MAP-based blur kernel estimation model in the gra dient space,…”

“…To enforce sparsity on k, in the implemen tations [4,IS,29,21] the hard-thresholding operator was adopted to make ki � c, 'Vi, where c is some small pos itive value. Other sparsity priors, e.g., TV [3] and hyper Laplacian [14,13], were also suggested to avoid the non blur solution and converge to desired solution [24].…”