|||
Wavelet Image Inpainting Based on
Dictionary Learning with a Beta Process
Abstract - The problem of image inpainting and wavelet image inpainting were presented in this study.
Dictionary Learning with a Beta Process (BPDL) was introduced. A new method based on BPDL
was proposed for wavelet image inpainting. Unlike conventional methods which mostly based on diffusion
theory in physics, this method is based on sparse image representation and considers an image as a
combination of different structural patterns to achieve inpainting. The image simulation experiments
were designed to test the algorithm. The results demonstrated that the connectivity principle of human
perception was well realized with good vision effect. The PSNR of the wavelet coefficients partly damaged
images was improved significantly after processed by the new method. It's also available for NIR images.
It's concluded that the presented method based on BPDL was an effective method for wavelet image
inpainting.
Keywords - Image Inpainting, Wavelet, Dictionary Learning, Beta process
Archiver|手机版|科学网 ( 京ICP备07017567号-12 )
GMT+8, 2024-12-22 01:06
Powered by ScienceNet.cn
Copyright © 2007- 中国科学报社