Deep Image Harmonization

Abstract

Compositing is one of the most common operations in photo editing. To generate realistic composites, the appearances of foreground and background need to be adjusted to make them compatible. Previous approaches to harmonize composites have focused on learning statistical relationships between hand-crafted appearance features of the foreground and background, which is unreliable especially when the contents in the two layers are vastly different. In this work, we propose an end-to-end deep convolutional neural network for image harmonization, which can capture both the context and semantic information of the composite images during harmonization. We also introduce an efficient way to collect large-scale and high-quality training data that can facilitate the training process. Experiments on the synthesized dataset and real composite images show that the proposed network outperforms previous state-of-the-art methods.

References in corpus (8)

• Caffe: Convolutional Architecture for Fast Feature Embedding
• Semantic Image Segmentation with Deep Convolutional Nets and Fully Connected CRFs
• Fast and Accurate Deep Network Learning by Exponential Linear Units (ELUs)
• Fully Convolutional Networks for Semantic Segmentation
• Semantic Understanding of Scenes through the ADE20K Dataset
• Learning Representations for Automatic Colorization
• Photo Aesthetics Ranking Network with Attributes and Content Adaptation
• Learning a Discriminative Model for the Perception of Realism in Composite Images

Cited by in corpus (13)

• GP-GAN: Towards Realistic High-Resolution Image Blending
• Gated Fusion Network for Single Image Dehazing
• ST-GAN: Spatial Transformer Generative Adversarial Networks for Image Compositing
• Image Inpainting using Block-wise Procedural Training with Annealed Adversarial Counterpart
• OPA: Object Placement Assessment Dataset
• Semi-parametric Image Synthesis
• Image Correction via Deep Reciprocating HDR Transformation
• FishEyeRecNet: A Multi-Context Collaborative Deep Network for Fisheye Image Rectification
• Where and Who? Automatic Semantic-Aware Person Composition
• Temporally Coherent Video Harmonization Using Adversarial Networks
• HYouTube: Video Harmonization Dataset
• Deep Image Compositing
• Deep Semantics-Aware Photo Adjustment