Automated Deep Alpha Matting for Vehicle Images

车辆图像的自动深度 Alpha 抠图

• 批准号: 523064-2018
• 负责人: Chen, Jun
• 金额: $ 1.82万
• 依托单位: McMaster University
• 依托单位国家: 加拿大
• 项目类别: Engage Grants Program
• 财政年份: 2018
• 资助国家: 加拿大
• 起止时间: 2018-01-01 至 2019-12-31
• 项目状态: 已结题
• 来源: https://www.nserc-crsng.gc.ca/ase-oro/Details-Detailles_eng.asp?id=647352
• 关键词: Automated; Deep; Alpha; Matting; Vehicle

Current methods for image matting are primarily based on manual boundary division and judgement, which**requires skilled editors and costs enormous amounts of time. More importantly, since the acquired images are**obtained by the human driven process of manual division, the process often results in errors within the**boundaries and detailed sections, which produces unnatural re-composited images. Alpha matting is a proposed**solution, which uses automatic technology to replace the manually based boundary division. By incorporating**the progress of deep learning technology in recent years with alpha matting, this yields higher accuracy as well**as wider application use cases. The proposed project will help to build the foundation of a long-term**collaboration between McMaster University and Car Media 2.0, which is a modern business-to-business**vehicle media provider that captures and edits photos of vehicles for the purposes of advertising throughout**US, Canada and Europe. It is anticipated that the use of this improved high-precision image matting system**will yield image matting processes that are more efficient, more accurate, and less costly. The competitive**advantage gained by this technology will help Car Media 2.0 to further grow and expand its business, thus**creating new employment opportunities in Canada. This project will also provide the participating students**with rich training experiences in deep learning, which is becoming an enabling technology in the big data era.

目前的图像遮片方法主要是基于人工边界划分和判断，这**需要熟练的编辑人员，并且花费大量的时间。更重要的是，由于获取的图像是**由人工驱动的手动分割过程获得的，该过程经常导致**边界和细节部分的错误，从而产生不自然的重新合成图像。Alpha Matting是一种建议的**解决方案，它使用自动技术来取代基于手动的边界划分。通过将**近年来深度学习技术的进步与Alpha Matting相结合，这将产生更高的精确度**以及更广泛的应用用例。拟议中的项目将有助于为麦克马斯特大学和汽车媒体2.0之间的长期**合作奠定基础，后者是一家现代企业对企业**车辆媒体提供商，为**美国、加拿大和欧洲的广告目的捕捉和编辑车辆的照片。预计这种改进的高精度图像遮片系统**的使用将产生更高效、更准确、成本更低的图像遮片过程。这项技术获得的竞争优势将帮助Car Media 2.0进一步发展和扩大其业务，从而**在加拿大创造新的就业机会。该项目还将为参与的学生提供丰富的深度学习培训经验**，深度学习正在成为大数据时代的使能技术。