Three-dimensional shape-adaptive wavelet transform for efficient object-based video coding

摘要

A three-dimensional (3D) shape-adaptive discrete wavelet transform (SA-DWT) is provided for efficient object-based video coding. In a first stage, a one-dimensional SA-DWT is performed along the temporal direction among pixels that have temporal correspondence. The correspondence can be established by motion estimation or other matching approaches. SA-DWT in the temporal direction is used to treat emerging pixels, terminating pixels or pixels that have colliding correspondence pixels. After the temporal SA-DWT transform, the resulting temporal wavelet coefficients are placed in the spatial positions corresponding to the original pixels to maintain the spatial correlation within each frame. Then, in a second stage, a two-dimensional SA-DWT is applied to the temporal SA-DWT coefficients within each frame. The 3D SA-DWT can handle arbitrarily shaped video objects while providing flexible spatial and temporal scalability as in any wavelet-based coding scheme. The 3D SA-DWT can also track the video object motion and perform the wavelet transform among corresponding pixels for that object while keeping the spatial correlation within a frame.