Non-negative scaling functions and applications in image coding

At medium and low bit rates, traditional subband image coding algorithms suffer from basis related visual artifacts. Short basis causes blocking and long basis causes ringing. In particular, ringing is caused by the undershoot and overshoot of time domain response of the basis functions. They can be effectively suppressed by using a scaling function that monotonically decreases to zero at block edges. In this paper, the theory, structure and design of filter banks with nonnegative scaling function are analyzed and their application in image coding is investigated. Experiments show that both blocking and ringing noises can be effectively suppressed by using the proposed filter banks. Moreover, they can be parametrized by lattice structures.