Inharmonious Region Localization by Magnifying Domain Discrepancy
This addresses the issue of detecting visual inconsistencies in edited images for applications like photo editing and forensics, but it is incremental as it builds on existing localization networks with a novel loss.
The paper tackles the problem of localizing inharmonious regions in synthetic images caused by color and illumination inconsistencies from editing, by transforming images to magnify domain discrepancies, resulting in superior performance on an image harmonization dataset.
Inharmonious region localization aims to localize the region in a synthetic image which is incompatible with surrounding background. The inharmony issue is mainly attributed to the color and illumination inconsistency produced by image editing techniques. In this work, we tend to transform the input image to another color space to magnify the domain discrepancy between inharmonious region and background, so that the model can identify the inharmonious region more easily. To this end, we present a novel framework consisting of a color mapping module and an inharmonious region localization network, in which the former is equipped with a novel domain discrepancy magnification loss and the latter could be an arbitrary localization network. Extensive experiments on image harmonization dataset show the superiority of our designed framework. Our code is available at https://github.com/bcmi/MadisNet-Inharmonious-Region-Localization.