P. Haavisto

Vector median filters

Two nonlinear algorithms for processing vector-valued signals are introduced. The algorithms, called vector median operations, are derived from two multidimensional probability density functions using the maximum-likelihood-estimate approach. The underlying probability densities are exponential, and the resulting operations have properties very similar to those of the median filter. In the vector median approach, the samples of the vector-valued input signal are processed as vectors. The operation inherently utilizes the correlation between the signal components, giving the filters some desirable properties. General properties as well as the root signals of the vector median filters are studied. The vector median operation is combined with linear filtering, resulting in filters with improved noise attenuation and filters with very good edge response. An efficient algorithm for implementing long vector median filters is presented. The noise attenuation of the filters is discussed, and an application to velocity filtering is shown. >

Post filtering methods for reducing blocking effects from coded images

Computationally simple post-filtering methods for block-coded image sequences are presented. Post-filtering design requires compromises between smoothing the block edges and retaining the image sharpness. The proposed filters offer a feasible method to improve the subjective quality of image sequences coded at low bit rates. >

Motion adaptive scan rate up-conversion

The interlaced scan format and the low frame rate in current television systems cause visible degradation in picture quality. To improve the picture, scan rate up-conversion can be implemented in the receiver. Typically, the up-conversion algorithms needed in stationary scenes are different from those needed in nonstationary scenes. In this paper we discuss problems related to scan rate up conversion and motion detection. We present an algorithm that adapts to the motion in the picture and yet effectively eliminates most of the artifacts caused by imperfect motion detection. The algorithm is based on a weighted median filter structure and a simple motion detector. The weights of the median filter are adapted according to the motion detector output. All algorithms discussed have been tested with real sequences using a video sequencer.

Fractional frame rate up-conversion using weighted median filters

The authors present algorithms for increasing the frame rate of an interlaced television system by a factor of 1.5. The algorithms utilize a weighted median filter structure and a simple motion detector. By adjusting the weights among the incoming fields the algorithms can be used with noninteger conversion factors. The motion detector guarantees that the performance with still pictures is optimal. The weighted median filter smoothly follows horizontal, vertical, and diagonal movement. A video sequencer was used to compare different interpolation algorithms. The proposed algorithms left few visible artifacts in the picture, and the improvement over algorithms without motion detection was particularly pronounced. The computational complexity and memory requirements of the algorithms are low. >

MEDIAN BASED IDEMPOTENT FILTERS

Idempotent filters produce a root signal in a single filter pass, i.e. the filter output is invariant to further filterings with the same filter. In this paper median based idempotent filter structures are introduced. Two approaches to generate these filters are studied: weighted median filters and median filter cascades. Two subclasses of n-dimensional idempotent weighted median filters, called Class 1 and Class 2 filters in the paper, are introduced. It is shown that both Class 1 and Class 2 filters suppress impulsive noise from n-dimensional input signals and yet have almost no effect on the non-corrupted parts of the signal. These filters are therefore well-suited for example for preprocessing purposes. An application to speech processing is described. Other likely applications of these filters are in image processing and, also, in image sequence processing, where the filter mask is typically 3-dimensional. Sufficient conditions for a filter cascade to be idempotent are given. Two idempotent median filter cascades and their advantages are discussed.

On VLSI implementation of median based field rate up-conversion

Field rate up-conversion using adaptive weighted median filtering is described. The weights of the algorithm are adapted according to the output from a motion detector. High-speed VLSI circuit architectures for the computation of weighted median are discussed. A bit-sliced architecture based on a previously introduced bit-serial method, having good expandability and supporting computation of weighted medians, is described.<<ETX>>

Motion Compensated Deinterlacing

A new algorithm to convert interlaced video sequences to a progressive format is proposed. The algorithm is based on motion detection, simple motion compensation, and adaptive weighted median filtering. The weights are adapted to emphasize temporal information in stationary areas and spatial information in the moving areas. Compensation of commonly occuring horizontal motion is used to improve the performance without adding the extra complexity of full motion compensation. A novel improvement to direction dependent processing is used to enchance the interpolation result on diagonal edges. The method can be used in advanced television receivers.

Signal Processing Based on Maximum Likelihood Selection and Linear Filtering

In this paper we describe a class of non-linear digital maximum likelihood filters that consist of a linear system and a selection element. The data is input to the linear element which has several outputs. The selection element chooses one of these to be the final filter output. The output is chosen so that assuming an exponential input distribution it is the sample most likely to be the correct signal level. The median filter is a special case in this filter class.Selection rules are derived for both scalar and multispectral samples. In the multispectral case the concept of vector median filters is described. The structure and purpose of the linear system is discussed. The root signals of the filters are studied. Examples of the filter performance are given and applications to image processing are shown in the areas of image smoothing and coding and edge detection.

Generating idempotent filters

Median based idempotent filter structures are introduced. Two approaches to generate these filters are studied: weighted median filters and median filter cascades. Two subclasses of n-dimensional idempotent weighted median filters, called Class 1 and Class 2 filters, are introduced. It is shown that both Class 1 and Class 2 filters suppress impulsive noise from n-dimensional input signals and yet have almost no effect on the noncorrupted parts of the signal. These filters are therefore well suited for preprocessing purposes. Likely applications of these filters are in speech processing, image processing, and image sequence processing, where the filter mask is typically three-dimensional. Sufficient conditions for a filter cascade to be idempotent are given. Two idempotent median filter cascades and their advantages are discussed.<<ETX>>