Deformable Kernel Networks for Joint Image Filtering

TL;DR

This paper introduces Deformable Kernel Networks (DKN), a novel CNN-based approach that learns spatially-variant kernels for joint image filtering, outperforming previous methods in various image enhancement tasks.

Contribution

We propose a new CNN architecture, DKN, that adaptively learns sparse, spatially-variant kernels for improved joint image filtering, with a fast implementation for practical use.

Findings

Outperforms state-of-the-art in multiple image filtering tasks

Fast DKN runs 17 times faster on standard images

Sparse 3x3 kernels achieve superior filtering results

Abstract

Joint image filters are used to transfer structural details from a guidance picture used as a prior to a target image, in tasks such as enhancing spatial resolution and suppressing noise. Previous methods based on convolutional neural networks (CNNs) combine nonlinear activations of spatially-invariant kernels to estimate structural details and regress the filtering result. In this paper, we instead learn explicitly sparse and spatially-variant kernels. We propose a CNN architecture and its efficient implementation, called the deformable kernel network (DKN), that outputs sets of neighbors and the corresponding weights adaptively for each pixel. The filtering result is then computed as a weighted average. We also propose a fast version of DKN that runs about seventeen times faster for an image of size 640 x 480. We demonstrate the effectiveness and flexibility of our models on the tasks of depth map upsampling, saliency map upsampling, cross-modality image restoration, texture removal, and semantic segmentation. In particular, we show that the weighted averaging process with sparsely sampled 3 x 3 kernels outperforms the state of the art by a significant margin in all cases.