Iterative image coding using hybrid wavelet-based triangulation

This paper presents a novel hierarchical subdivision scheme for generating triangular meshes to represent intensity images. The resulting representation is well suited to content-based management of images and videos, including storage and retrieval for multimedia applications. Mesh node locations are selected using multiresolution wavelet-based analysis, and this leads to a piecewise-linear approximation of image intensity values. Interpretation of wavelet coefficients at successively finer levels leads to fast, efficient triangular mesh construction in which the number of approximating elements depends on the image content and on user-specified maximum-error. By combining Delaunay and data-dependent triangulation approaches, the result is an efficient multiresolution coding scheme that approximates the image to a desired level of detail, while retaining information related to image content under user-specified triangulation regularity. We demonstrate the utility of this approach for such multimedia applications such as selective compression, segmentation, and eye detection in images of human faces.