Parallel (or block) FIR digital lters can be used either for high-speed or low-power (with reduced supply voltage) applications. Traditional parallel lter implementations cause linear increase in the hardware cost with respect to the block size. Recently, an eecient parallel FIR lter implementation technique requiring a less-than linear increase in the hardware cost was proposed. This paper mak...

In this paper, methods for processing arbitrary-length input signals with MDFT lter banks will be presented. The MDFT lter bank can be regarded as the most general type of a special class of lter banks. These lter banks have linear phase analysis lters, but di erent centers of symmetry due to subsampling with and without a phase shift [1], [2]. Already known extension methods cannot be applied ...

The classical method for impulsive noise removal in image signals consists of two steps, i.e., a step for impulse detection and a step for estimation of the pixels corrupted by impulsive noise. In this paper, we show that the step of impulse detection can be described very suitably by piecewise-linear (PWL) functions which explains intuitively the process of detection as a partition of the sign...

A cubic polynomial lter is presented for edge preserving noise smoothing. The lter coeecients are adaptable and can be optimized to improve the l-ter result. Experiments show that the image quality achieved with the optimized cubic lter is comparable to that of a more complex rational lter.

The performance of subband image coders depends on proper choice of lter banks. Although odd length lters in the lter banks produce waveform type artifacts, this can be alleviated by enforcing a smooth interpolation property to the synthesis lowpass lter. Evaluation of even, odd, and combinations of even and odd length lters in tree-structured lter banks, where the lter coe cients are obtained ...

Modern high energy physics experiments have to process terabytes of input data produced in particle collisions. The core of the data reconstruction in high energy physics is the Kalman lter. Therefore, the speed of Kalman lter based algorithms is of crucial importance in on-line data processing. This is especially true for the combinatorial track nding stage where the Kalman lter based track t ...

We've found that Burg's two-dimensional lter ooers too much freedom in producing lateral predictions, since the predictions within a trace appear to overwhelm the lateral predictions. Instead, we x the lter coeecients along the output column so the form of the two-dimensional lter becomes equivalent to that of the FX-decon operator. The ability to control the length of this lter in time allows ...

A novel Fast RLS Algorithm based on the Givens Rotation and developed from an UDU T square-root fac-torization of autocorrelation matrix is discussed. The algorithm presents excellent numerical properties and requires 14N multiplications and 6N divisions per sampling interval, where N is the linear lter order.

A popular technique for time delay estimation is to use an FIR #lter to model the time di-erence and the #lter weights are interpolated with a sinc function to obtain the delay estimate. However, the sinc interpolator requires a su/ciently long #lter length for accurate delay estimation. In this paper, we propose to process the #lter weights via a least-squares-based method in order to acquire ...

We present a novel coupled £lter pair for denoising, segmentation, and inpainting, based on the well known MumfordShah image model. Our self-contained digital extension to a general graph domain makes it possible to cheaply and universally bene£t from this conventionally expensive yet very powerful image model.