Signal Processing: Image Communication 65 (2018) 141–153 Contents lists available at ScienceDirect Signal Processing: Image Communication journal homepage: www.elsevier.com/locate/image Multi-purpose bi-local CAT-based guidance filter Hristina Hristova *, Olivier Le Meur, Remi Cozot, Kadi Bouatouch University of Rennes 1, 263 Avenue General Leclerc, Rennes, France ARTICLE INFO Keywords: Guidance filter Chromatic adaptation transform Image filtering ABSTRACT In this paper, we propose a new guidance filter, based on color perception through a chromatic adaptation model. Our method consists of a patch-wise linear transformation, which transfers details from a guidance image to an input image. The amount of transferred details is controlled by a novel chromatic adaptation transform (CAT), called bi-local CAT, embedded in our method. The bi-local CAT contributes to the detail recovery from the guidance image as well as to the preservation of the input reflections and shadows. Our bi-local CAT-based guidance filter is applied in various image processing domains, such as image denoising, image deblurring, texture transfer, detail enhancement, skin beautification, etc. 1. Introduction Many image processing applications require image smoothing tech- niques for noise reduction. Classical filters, such as Gaussian filter and median filter, often blur edges in the smoothing process. Edge- preserving filters like the bilateral filter [1] also experience a trade-off between the noise removal and the image integrity, as they use a single image to build their kernel. In contrast, guidance-based filters incorporate additional informa- tion into the filtering process through the use of a guidance image. The guidance image, which is often a noise-free sharp image, is used explicitly in the estimation of the filter kernel. That way, the smoothing of the input image is carried out more efficiently and the amount of information loss is reduced. The guidance-based filters are commonly applied for example-based noise reduction [2–4], depth-map filter- ing [5], image matting [2], etc. Local optimizations in the guidance-based filters may concentrate blur around sharp edges and cause a decrease in the sharpness of the output image. The guidance-based filters may compromise the input lighting atmosphere by smoothing down input reflections. Furthermore, they do not recover details from the guidance image, which limits their applicability to texture transfer and detail enhancement. To tackle some of the limitations of the guidance-based filters, we present a new guidance filter, based on color appearance and color perception [6]. Our filter carries out a patch-wise linear transformation between an input image and a guidance image. In practice, this linear transformation adds details from the guidance image to a low-pass version of the input image. The amount of added details is controlled by a scaling coefficient in which we embed a new chromatic adaptation * Corresponding author. E-mail address: hristinaih@gmail.com (H. Hristova). transform (CAT), called bi-local CAT [7]. The bi-local CAT strongly contributes to preserving input scene details such as input reflections. Along with preserving the input scene ambience, the embedded bi-local CAT plays a major role in transferring details from the guidance image. In comparison to existing guidance-based filters, our bi-local CAT- based guidance filter provides solutions to various digital imaging prob- lems. In this paper, we address several image processing applications, i.e. image denoising, texture transfer, detail enhancement with near- to-infrared (NIR) images, image deblurring, mask refinement and skin beautification. Our results compare fairly to results from state-of-the-art methods. Furthermore, our filter outperforms existing guidance-based filters in terms of image sharpness, detail enhancement and preservation of the input lighting. The rest of the paper is organized as follows. Existing guidance- based filters along with their advantages and limitations are discussed in Section 2. Section 3 introduces our bi-local CAT-based guidance filter. Applications and results are presented in Section 4. Finally, the last section concludes the paper. 2. Related work In the following section, we present important guidance-based image filters and we discuss their objectives and functionalities as well as their limitations. The input of the guidance-based filters consists of an input image  and a guidance image . The main goal of guidance-based filters is to filter the input image  using information from the guidance . Depending on how the information of  is incorporated into the filtering process, the guidance-based filters can be classified into two categories: https://doi.org/10.1016/j.image.2018.03.010 Received 13 January 2018; Received in revised form 17 March 2018; Accepted 17 March 2018 Available online 8 April 2018 0923-5965/© 2018 Elsevier B.V. All rights reserved.