Three-dimensional wavelet coding of video with arbitrary regions of support

We present a method for the compression of images and video with arbitrary regions of support. We start by constructing a wavelet transform which adapts to an arbitrarily shaped image. We then modify Said and Pearlman's Set Partitioning in Hierarchical Trees (SPIHT) algorithm to efficiently code wavelet coefficients of arbitarily shaped regions. The wavelet transform for arbitrary region of support and the modified SPIHT algorithm are extended to 3-D to encode video sequences of arbitrary regions of support. The combination of these two techniques achieves gains of 5.60 dB over a rectangular 3-D wavelet transform used in conjunction with the original 3-D SPIHT coding.