PROJECT TITLE :

A Benchmark for Edge-Preserving Image Smoothing

ABSTRACT:

An key step in many low-level vision challenges is edge-preserving image smoothing. Algorithms have been proposed, however they face a number of challenges in their implementation. Because most existing algorithms only have a few parameter settings, they can't effectively handle a large variety of image content types. A dearth of widely acknowledged datasets for objectively comparing edge-preserving picture smoothing methods makes performance evaluation difficult. In this research, we propose a benchmark for edge-preserving image smoothing in order to solve these challenges and further improve the state of the art. It comprises an image dataset with ground truth smoothing results and baseline techniques that can produce competitive edge-preserving smoothing results for a wide variety of image contents.. While the established dataset contains 500 training and testing images that represent a wide range of visual object categories, the baseline methods in our benchmark are based on representative deep convolutional network architectures, on top of which we design novel loss functions that are well suited for the edge-preserving image smoothing With leading outcomes both intuitively and numerically, the trained deep networks outperform most of the current state-of-the-art methods. The metric will be made available to the general audience.