Abstract | ||
---|---|---|
Advertisement logo compositing is aiming to embed some specified logos in a suitable position of target images with alike geometric distortion and same appearance characteristic. Unsupervised learning has gained considerable attention recently. To synthetically address the problems about geometric distortion and appearance realism, we propose a novel learnable module, the adversarial geometric consistency pursuit model (AGCP), which explicitly allows seamless image compositing. On one hand, we design an adversarial structure that generates composite images taking geometric correction and appearance harmonization into account. Our proposed adversarial learning approach is able to obtain better harmonization in the region of interest. On the other hand, a novel geometric consistency pursuit loss is designed which encourages the network to learn the warp parameters of target images while preserving the feature of the source object. Our comparative evaluation demonstrates the effectiveness of the proposed method. |
Year | DOI | Venue |
---|---|---|
2019 | 10.1109/VCIP47243.2019.8965885 | 2019 IEEE Visual Communications and Image Processing (VCIP) |
Keywords | Field | DocType |
advertisement logo compositing,adversarial learning,geometric consistency pursuit,seamless synthesis,projective distortion | Computer vision,Advertising,Computer science,Logo,Unsupervised learning,Geometric consistency,Artificial intelligence,Region of interest,Distortion,Compositing,Adversarial system | Conference |
ISSN | ISBN | Citations |
1018-8770 | 978-1-7281-3724-7 | 0 |
PageRank | References | Authors |
0.34 | 2 | 5 |
Name | Order | Citations | PageRank |
---|---|---|---|
Xiang Li | 1 | 0 | 0.34 |
Guowei Teng | 2 | 0 | 0.34 |
Ping An | 3 | 545 | 68.73 |
Haiyan Yao | 4 | 3 | 2.75 |
Yilei Chen | 5 | 0 | 0.34 |