Deep Joint Image Filtering
Joint image filters can leverage the guidance image as a prior and transfer the structural details from the guidance image to the target image for suppressing noise or enhancing spatial resolution. Existing methods rely on various kinds of explicit filter
- PDF / 6,792,519 Bytes
- 16 Pages / 439.37 x 666.142 pts Page_size
- 96 Downloads / 205 Views
University of California, Merced, Merced, USA {yli62,mhyang}@ucmerced.edu 2 University of Illinois, Urbana-Champaign, Champaign, USA {jbhuang1,n-ahuja}@illinois.edu https://sites.google.com/site/yijunlimaverick/deepjointfilter
Abstract. Joint image filters can leverage the guidance image as a prior and transfer the structural details from the guidance image to the target image for suppressing noise or enhancing spatial resolution. Existing methods rely on various kinds of explicit filter construction or handdesigned objective functions. It is thus difficult to understand, improve, and accelerate them in a coherent framework. In this paper, we propose a learning-based approach to construct a joint filter based on Convolutional Neural Networks. In contrast to existing methods that consider only the guidance image, our method can selectively transfer salient structures that are consistent in both guidance and target images. We show that the model trained on a certain type of data, e.g., RGB and depth images, generalizes well for other modalities, e.g., Flash/Non-Flash and RGB/NIR images. We validate the effectiveness of the proposed joint filter through extensive comparisons with state-of-the-art methods. Keywords: Joint filtering
1
· Deep convolutional neural networks
Introduction
Image filtering with a guidance signal, known as joint or guided filtering, has been successfully applied to a variety of computer vision and computer graphics tasks, such as depth map enhancement [1–3], joint upsampling [1,4], cross-modality noise reduction [5–7], and structure-texture separation [8,9]. The wide applicability of joint filters can be attributed to their adaptability in handling visual signals in various visual domains and modalities, as shown in Fig. 1. For a target image, the guidance image can either be the target image itself [6,10], high-resolution RGB images [2,3,6], images from different sensing modalities [5,11,12], or filtering outputs from previous iterations [9]. The basic idea behind joint image filtering is that the guidance image often contains important structural details that can be transferred to the target image. The main goal of joint filtering is to enhance the degraded target image due to noise or low spatial resolution while Electronic supplementary material The online version of this chapter (doi:10. 1007/978-3-319-46493-0 10) contains supplementary material, which is available to authorized users. c Springer International Publishing AG 2016 B. Leibe et al. (Eds.): ECCV 2016, Part IV, LNCS 9908, pp. 154–169, 2016. DOI: 10.1007/978-3-319-46493-0 10
Deep Joint Image Filtering
Depth upsampling
Noise reduction
Inverse halftoning
155
Texture removal
Fig. 1. Sample applications of joint image filtering: depth map upsampling, cross-modal noise reduction (flash/non-flash), inverse halftoning, and edge-preserving smoothing for texture removal. The Target/Guidance pair (top) can be various types of crossmodality visual data. With the help of the guidance image, important structures can be transferred to the degrad
Data Loading...
