Edge-based image interpolation using symmetric biorthogonal wavelet transform
Weizhong Su · cIRcle (University of British Columbia) · 2009
Image interpolation is an important part of digital image processing. Many approaches are proposed for enlarging and reducing images. Recently, most papers in image interpolation are focused on edge-based interpolation since sharp edges and smooth contours can give better impression to the human vision than others. L i & Orchard and Kimmel proposed edge-based interpolation approaches that can produce better image quality compared with the traditional methods such as bilinear and bicubic interpolations. In this thesis, a new edge-based image interpolation approach that uses symmetric biorthogonal wavelet transforms is proposed. According to wavelet multiresolution analysis theory, an image can be decomposed into a series of approximation sub-images and detail sub-images with horizontal, vertical, and diagonal edge information. Based on this theory, many wavelet-based interpolation approaches have been proposed. However, most of them are computationally expensive or not efficient. In this thesis, we set up a list of ideal step edge models, and explore the relationships between the wavelet approximation sub-image and the three wavelet detail sub-images of these models. Based on these relationships, a fast and efficient algorithm that predicts the edge information of the interpolated image is proposed. The results of our experiments prove that the wavelet-based image interpolation with our new approach has good performance compared with other state-ofthe- art image interpolation approaches. In conclusion, the 9 / 7 -M inverse wavelet transform with our new approach is the best solution for image interpolation.