Numerous mobile videos get recorded in wide-angle mode through lenses. The videos include the usage of narrative subjects and background. However, the problem arises with video playbacks as they display video distortions that lead to stretching of the subjects surrounding the image corners.
The situation might be solved with the help of a new study that addresses the problem of wide-angle face distortion on videos that use spatial-temporal mesh optimization. The developed algorithm generates a warping mesh to change as per the stereographic projection present on the facial region. The team also introduced coherent embedding terms and temporal smoothness to maintain the temporal consistency of meshes. Furthermore, the addition of a line-preservation term also helps preserve straight lines in the background. The algorithm could significantly contribute to Digital Video Market as it helps create a video with natural-looking human faces.
The researchers collected a benchmark dataset of videos through cameras present within iPhone 11, Google Pixel 3 and GoPro. This was done for the benefit of performance evaluation. The study demonstrated that the new approach enhances the video quality and is also suitable for post-editing software.
Taking video blogs and selfies are a popular way of interacting with social media for people. These formats are often captured at wide-angle cameras either to show human subjects or to obtain expanded background. However, because of the perspective projection, faces at the corners of the frame's edges exhibit distortions due to which facial features get stretched or swished. This results in people creating a video of poor quality that is unpleasant to the eye. Thus, to overcome this problem, researchers brought forth a video wrapping algorithm that can correct distortions.
Their primary idea was to apply stereographic projection at the site of facial regions. The team formulated a mesh warp problem to utilize minimized background deformation and spatial-temporal energy minimization to achieve this. While undertaking this task, researchers also wanted to maintain the straight-edges in the background, which was done by line preservation.
Further, the study also addresses temporal coherency. The team restricted the temporal smoothness upon the warping mesh and facial trajectories with the help of latent variables. The researchers also evaluated the performance of their development. For this, they developed a wide-angle video dataset with an extensive range of focal lengths. Their study and evaluations showed that 83.9 per cent of users showed preference to the novel algorithm compared to other alternatives. The results denote immense potential in the new algorithm, and it could significantly improve the digital video quality.
