Least‐mean‐squares design of filters controlling subband coefficient powers with applications to image coding
Hiroshi Ito · Electronics and Communications in Japan (Part I Communications) · 2002
Abstract In image coding, the relationship between the coding rate and distortion depends on the input image. In coding that decomposes a signal into multiple frequency bands (subbands) as in the DCT and the wavelet transform, the power distribution of the decomposed signal (subband coefficients) is the primary factor that determines the rate‐distortion characteristics. In this paper, we propose a design method for prefilters that controls the powers of the subband coefficients. This design method assigns a target gain for the prefilter to each subband and determines the filter coefficients by least mean squares estimation. The input signal is used in the least mean squares calculation. Thus, if this filter is used in the stage prior to coding, the target rate distortion characteristics adapted to the input signal can be more faithfully achieved. By smoothly assigning the target gains to adjacent subbands, this prefilter can accurately achieve the target gains even for a low‐order analysis filter as in DCT. In addition, computer simulations demonstrated satisfactory performance when applied to image coding. © 2002 Scripta Technica, Electron Comm Jpn Pt 1, 85(5): 11–24, 2002