Deep Feature Consistent Deep Image Transformations: Downscaling, Decolorization and HDR Tone Mapping
Abstract
Building on crucial insights into the determining factors of the visual integrity of an image and the property of deep convolutional neural network (CNN), we have developed the Deep Feature Consistent Deep Image Transformation (DFC-DIT) framework which unifies challenging one-to-many mapping image processing problems such as image downscaling, decolorization (colour to grayscale conversion) and high dynamic range (HDR) image tone mapping. We train one CNN as a non-linear mapper to transform an input image to an output image following what we term the deep feature consistency principle which is enforced through another pretrained and fixed deep CNN. This is the first work that uses deep learning to solve and unify these three common image processing tasks. We present experimental results to demonstrate the effectiveness of the DFC-DIT technique and its state of the art performances.
Overview
We seek to train a convolutional neural network as a non-linear mapper to transform an input image to an output image following what we call the deep feature consistent principle. Our system consists of two components: a transformation network and a loss network. A convolutional neural network transforms an input to an output and a pretrained deep CNN is used to compute feature perceptual loss for the training of the transformation network.
Results
We provide the comparison results for image downscaling, decolorization and HDR image tone mapping. Additional results trained with different level feature perceptual loss are also provided for each task.
Click here for more results
A comparison with different level feature perceptual loss.
A comparison with different level feature perceptual loss.
A comparison with different level feature perceptual loss.
A demonstration of the effects of logarithmic compression.
