Evaluating Parameterization Methods for Convolutional Neural Network (CNN)-Based Image Operators

Seung-Wook Kim, Sung-Jin Cho, Kwang-Hyun Uhm, Seo-Won Ji, Sang-Won Lee, Sung-Jea Ko; Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) Workshops, 2019, pp. 0-0

Recently, deep neural networks have been widely used to approximate or improve image operators. In general, an image operator has some hyper-parameters that change its operating configurations, e.g., the strength of smoothing or up-scale factors in super-resolution. To address the varying parameter setting, an image operator taking such parameters as its input, namely a parameterized image operator, is an essential cue in image processing. Since many types of parameterization techniques exist, comparative analysis is required in the context of image processing. In this paper, we therefore analytically explore the operation principles of these parameterization techniques and study their differences. In addition, performance comparisons between image operators parameterized by using these methods are assessed experimentally on common image processing tasks including image smoothing, denoising, deblocking, and super-resolution.