Generating Pixel Art from Game Characters with Convolutional-Neural Network
1
조회
0
다운로드
0
인용
0
참고문헌
키워드
초록
Pixel art, which presents low-resolutional images with restricted color palette, has been employed frequently in the early computer games played on low memory capacity and computational performance. Recently, pixel art wides its applications to the area such as puzzle and game. In this paper, we present a pixel art generator from images of game characters. Unlike traditional framework, we employ and modify a Convolutional-Neural Network(CNN) to generate pixel art by placing an up-convolution layer after convolution layers. The up-convolution layer increases the resolution of the result images to satisfy user-required resolution. The network is trained by minimizing the Mean Squared Error(MSE) between ground truth images and generated pixel art images from the input high-resolutional image. Also, we employ artists to produce the ground truth of pixel art for our network and augment the data by rotating and fliping. We partition the ground truth images into three datasets: a training, validation and test dataset. With this dataset, we perform a supervised learning and train our network as the pixel art generator. We show a training process and a training accuracy. Moreover, we test our architecture for a various images as well as the test dataset to prove the excellence of our architecture.
Abstract
Pixel art, which presents low-resolutional images with restricted color palette, has been employed frequently in the early computer games played on low memory capacity and computational performance. Recently, pixel art wides its applications to the area such as puzzle and game. In this paper, we present a pixel art generator from images of game characters. Unlike traditional framework, we employ and modify a Convolutional-Neural Network(CNN) to generate pixel art by placing an up-convolution layer after convolution layers. The up-convolution layer increases the resolution of the result images to satisfy user-required resolution. The network is trained by minimizing the Mean Squared Error(MSE) between ground truth images and generated pixel art images from the input high-resolutional image. Also, we employ artists to produce the ground truth of pixel art for our network and augment the data by rotating and fliping. We partition the ground truth images into three datasets: a training, validation and test dataset. With this dataset, we perform a supervised learning and train our network as the pixel art generator. We show a training process and a training accuracy. Moreover, we test our architecture for a various images as well as the test dataset to prove the excellence of our architecture.
참고문헌 (17)
- T.C. Inglise, "Pixelating Vector Line Art", Proceedings of the Symposium on Non-Photorealistic Animation and Rendering, pp.21~28 (2012)
- T.C. Inglise, "Rastering and Antialiasing Vector Line Art in the Pixel Art Style", Proceedings of the Symposium on Non-Photorealistic Animation and Rendering, pp.25~32 (2013)
- Z. Hu, "Cartesian Resizing of Line Drawing Pictures for Pixel Line Arts", Journal of IEICE Transactions on Information and Systems, E97.D(4), pp.1008~1010 (2014)[DOI]
- M.R. Huang, "Pixel Art Color Palette Synthesis", Proceedings of Information Science and Applications, pp.327~334 (2015)
- T. Gerstner, "Pixelated Image Abstraction", Proceedings of the Symposium on Non-Photorealistic Animation and Rendering, pp.29~36 (2012)
- T. Gerstner, "Pixelated Image Abstraction with Integrated User Constraints", Journal of Comput. Graph., 37(5), pp.333~347 (2012)
- M.H. Kuo, "PIXEL2BRICK: Constructing Brick Sculptures from Pixel Art", Journal of Computer Graphics Forum, 34(7), pp.339~348 (2015)
- M.H. Kuo, "Feature-aware Pixel Art Animation", Proceedings of the 24th Pacific Conference on Computer Graphics and Applications, pp.411~420 (2016)
- J. Kopf, "Depixelizing Pixel Art", Journal of ACM Trans. Graph., 30(4), pp.99~ (2011)
- Y. LeCun, "Gradient-based learning applied to document recognition", Journal of the IEEE, 86(11), pp.2278~2324 (1998)[DOI]
- P.O. Pinheiro, "From Image-Level to Puxel-Level Labeling With Convolutional Networks", Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition(CVPR), pp.1713~1721 (2015)
- J. Long, "Fully Convolutional Networks for Semantic Segmentation", Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition(CVPR), pp.3431~3440 (2015)
- J. Uhrig, "Pixel-Level Encoding and Depth Layering for Instance-Level Semantic Labeling", Proceedings of Pattern Recognition, pp.14~25 (2016)
- O. Ronneberger, "U-Net: Convolutional Networks for Biomedical Image Segmentation", Proceedings of Medical Image Computing and Computer-Assisted Intervention (MICCAI), pp.234~241 (2015)
- 양희경, "선택적 다중스케일 구조의 게임배경을 생성하는 영상 추상화 기법", 한국컴퓨터게임학회논문지, (19), pp.131~140 (2009), (사)한국컴퓨터게임학회
- 강동화, "CUDA를 이용한 실시간 감성 동영상 추상화 구현", 한국컴퓨터게임학회논문지, 24(3), pp.125~134 (2011), (사)한국컴퓨터게임학회
- 윤그림, "다중 스케일 복셀화에 기반한 표면 모델 추상화 기법", 한국컴퓨터게임학회논문지, 29(3), pp.1~8 (2016), (사)한국컴퓨터게임학회[DOI]