Smooth local sinusoidal bases on two-dimensional \(\mathbf L\)-shaped regions (Q1271502)

In object-based video coding, moving objects are frequently represented using a block DCT transform or a lapped orthogonal transform. However, these transforms have the restriction that they are only effective when the object is masked by a rectangular block. In this paper the author generalizes the lapped orthogonal transform and similar schemes based on local sinusoidal bases to non-rectangular regions, specifically L-shaped regions. A comparison of the signal-to-noise ratio between the two schemes is then made.