High resolution image restoration and image segmentation both require accurate characterization of singularity such as edges. The former one needs to enhance edges from low resolution images while the latter one needs to identify the locations of edges. Both become a challenging problem when the image contrast is very low, some boundary information is missing and when there is noise. We first review our algorithm of single image super resolution based on reproducible kernel Hilbert space and approximated Heaviside function. We then present our recent development on extension to 3D image super resolution. We also separate smooth components of an image from the edges using the same modeling and adopt for global and local image segmentation. Numerical results show the superiority of the proposed approaches.

会议网址：https://meeting.tencent.com/s/M5C8Zeg9nWTh

会议ID：621 108 179

会议密码：202011