Edge-preserving image filtering and applications


Student thesis: Doctoral Thesis

View graph of relations


  • Linchao BAO

Related Research Unit(s)


Awarding Institution
Award date16 Feb 2015


Edge-preserving image filtering has been serving as the foundation for many computer vision and graphics applications. Its traditional task is to smooth out noise/details in the image while preserving essential edges and structures presented in the input image, which could be very useful for image denoising, edge-aware image editing, or scene simplification for further analysis. The most well-known edge-preserving filtering algorithms include anisotropic diffusion and bilateral filtering. Despite its important role and wide use, edge-preserving filtering is still an active research area in both vision and graphics community. In this thesis, two novel edge-preserving filtering algorithms and related applications are proposed. Specifically, the relationship between edge-preserving filtering and a robust estimator (M-smoother) originated from statistics is first studied. Starting from the generalized formulation of M-smoother, a numerical scheme for solving it via a series of weighted-average filtering (e.g., box filtering, Gaussian filtering, bilateral filtering, and guided filtering) is proposed. Because of the equivalence between M-smoother and local-histogram-based filters (such as median filter and mode filter), the proposed framework enables fast approximation of histogram filters via a number of box filtering or Gaussian filtering. In addition, high-quality piecewise-constant smoothing can be achieved via a number of edge-preserving filtering integrated with the proposed framework. In the second part of the thesis, a novel edge-preserving filtering algorithm, namely "Tree Filtering", is proposed, exploring the 2D discrete signal space of an image. The proposed filter can smooth out high-contrast details while preserving major edges, which is not achievable for bilateral-filter-like techniques. Tree filter is a weighted-average filter, whose kernel is derived by viewing pixel affinity in a probabilistic framework simultaneously considering pixel spatial distance, color/intensity difference, as well as connectedness. Pixel connectedness is acquired by treating pixels as nodes in a minimum spanning tree (MST) extracted from the image. The fact that an MST makes all image pixels connected through the tree endues the filter with the power to smooth out high-contrast, fine-scale details while preserving major image structures, since pixels in small isolated region will be closely connected to surrounding majority pixels through the tree, while pixels inside large homogeneous region will be automatically dragged away from pixels outside the region. The tree filter can be separated into two other filters, both of which turn out to have fast algorithms. Thus the final algorithm is very efficient. In the third part of the thesis, a specific application relating to edge-preserving filtering, namely optical flow estimation, is studied. A novel large-displacement optical flow algorithm is proposed by applying edge-preserving idea into Patch-Match algorithm. The proposed algorithm is inspired by recent successes of edge-preserving filtering methods in visual correspondence searching as well as approximate nearest neighbor field algorithms. The main novelty is a fast randomized edge-preserving approximate nearest neighbor field algorithm which propagates self-similarity patterns in addition to offsets. Experimental results on public benchmarks show that the proposed method is significantly faster than state-of-the-art methods without compromising on quality, especially when scenes contain large-displacement motions.

    Research areas

  • Image processing, Computer vision