Wavelet image inpainting based on Dictionary Learning with a Beta Process

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.