一种图像背景虚化方法、装置、设备及存储介质
By using PortraitNet and Resnext50/U-Net networks for image segmentation and depth prediction, the real foreground and background are accurately segmented, solving the problem of poor background blur effect in smartphone cameras and achieving a more natural background blur effect.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- GUANGDONG OPPO MOBILE TELECOMMUNICATIONS CORP LTD
- Filing Date
- 2021-12-14
- Publication Date
- 2026-07-17
AI Technical Summary
In existing technologies, smartphone cameras have poor background blur effects due to their small aperture and short focal length. Furthermore, software simulation methods result in an unnatural transition between the foreground and background, and the blurred background lacks realism.
Image segmentation is performed using the Image Segmentation Network PortraitNet, combined with the monocular depth prediction network Resnext50 and the multi-scale depth fusion network U-Net. Accurate segmentation is achieved through image segmentation masks and depth images to determine the true foreground and background, and background blurring is then applied.
It improves the background blur effect, making the transition between the foreground and background more natural, and the blurred background effect is closer to the real blur effect.
Smart Images
Figure CN116266337B_ABST