Object-based video coding of arbitrarily shaped objects using lapped orthogonal transforms (LOTs) defined on rectangular and L-shaped regions

摘要

An object-based image/video coding algorithm that encodes arbitrary shaped objects using LOTs defined on rectangular and L-shaped regions. The regions of support of the blocks in the L-shaped regions are extended to form tapered overlapping regions that meet at the object's exterior and concave corners. The overlapping regions taper to the corners with a slope having a magnitude delta / epsilon , where delta and epsilon are the amounts of overlap between the blocks. This allows the basis and tapered window functions to be defined in the L-shaped region and specifically in the tapered overlapping regions in a manner that maintains the orthogonality of the basis functions while providing a smooth transition at the edges.