Image Feature Extraction and Reconstruction based on Filter Bank of Gaussian Derivative Wavelets

• Main Authors: Yung-Cheng Chen, 陳勇丞
• Other Authors: Shaw-Jyh Shin
• Format: Others
• Language: zh-TW
• Published: 2005
• Online Access: http://ndltd.ncl.edu.tw/handle/74002377110365806444

碩士 === 逢甲大學 === 通訊工程所 === 93 === Edge detection provides a representation that is used by all later stages of image processing. An edge representation must be perceptually complete. This means that a perceptually accurate estimate of the original image can be reconstructed from its edges. We present an algorithm for image reconstruction from edges using a redundant filter bank derived by a wavelet frame composed of derivatives of Gaussian functions. Edges of an image can be extracted by any edge detection method. The band-pass filtered image from this filter bank contains complete information of the original image. A two-stage scheme of image reconstruction from edges is established. First, the band-pass filtered image is approximated by a set of elementary functions specified by the initial edge parameters. The elementary functions comprise of the first three orders of Gaussian derivatives. Second, the optimally approximated band-pass filtered image plus the average of the image is applied to reconstruct the original image. We adopt a conjugate gradient algorithm for the reconstruction. Furthermore, the redundant filter bank is applied to irregular sampling of the original image. The sampled pixels are determined by the positions of valleys and ridges in the band-pass filtered image. The image can be reconstructed by an iterative interpolation method. The major contributions of this thesis are to lay down a theory of edge-based image reconstruction scheme and to provide an irregular sampling method.