Television – Image signal processing circuitry specific to television – Noise or undesired signal reduction
Reexamination Certificate
2001-03-14
2004-03-30
Miller, John (Department: 2614)
Television
Image signal processing circuitry specific to television
Noise or undesired signal reduction
C348S624000, C348S627000, C348S683000, C348S701000, C382S275000
Reexamination Certificate
active
06714258
ABSTRACT:
BACKGROUND OF THE INVENTION
1. Field of the Invention
The invention relates to a method and device for noise reduction.
2. Description of the Related Art
Noise reduction is generally known, and can be provided using a spatial or temporal noise reduction filter. For example, U.S. Pat. No. 5,400,083 describes a video-apparatus with a spatial noise reduction filter in the form of a vertical recursive noise reduction filter.
European Patent Application No. EP-A-0,497,222, corresponding to U.S. Pat. 5,119,195, discloses a video noise reduction system employing plural frequency bands, in which an input filter separates a luminance input signal into a high frequency component that is noise reduced by coring, and a low frequency component that is noise reduced by a frame recursive filter preceded by sub-sampling and followed by sample rate restoration by interpolation. An output circuit combines the noise reduced components to provide a processed output signal in which high frequency portions thereof are subjected to coring and low frequency portions thereof are subjected to recursive filtering. In a specific application, the frame recursive filter provides a plurality of low frequency sub-sampled components which are combined with the processed output signal in a further output circuit to provide a noise-reduced output signal in a progressive scan form.
European Patent Application No. EP-A-0,581,059 discloses a method of filtering noise in television or video signals by means of a noise reduction circuit having a first input which receives the input signal, and a second input which receives a low-frequency part of a field-delayed output signal of the noise reduction circuit. A decimation filter is present between an output of the noise reduction circuit and the field delay to reduce the data rate of the output signal of the noise reduction circuit. More specifically, the bandwidth is reduced by a factor 2, the data rate is reduced by the same factor, and the bit resolution is reduced from 8 bits to 7 bits. An interpolation filter is present between the field delay and the second input of the noise reduction circuit. The presence of the decimation filter and the interpolation filter allows the use of a field delay with a reduced storage capacity.
Noise reduction can be considered as an averaging process of the signal value of a pixel with that of neighboring pixels of which one has the confidence that they have approximately the same value as the first-mentioned pixel. Small differences are interpreted as noise and may be filtered. Large differences are assumed to be intended and must not be filtered. Spatial noise filtering uses spatial neighbors, i.e., the value of pixels in the immediate vicinity. Strong details will destroy the confidence in neighboring pixel values. Continued spatial filtering in the presence of details causes loss of sharpness, plastic faces, etc.
Video pictures are often quite static. Additive noise is usually random, moving and not static. This makes the noise quite obvious to the eye. By averaging a series of pictures, the picture content adds up and remains the same. The random noise content is uncorrelated and will be reduced according to the number of pictures averaged. Therefore, temporal noise reduction is applied. Temporal noise reduction will clean up the picture without affecting its resolution as long as the pictures are static. Temporal noise filtering uses temporal neighbors: the value of the same pixel in the past or future. The output signal Vo(n) of a known recursive temporal noise filter is a function of the output signal Vo(n−1) and the input video signal Vi(n) according to the relation:
Vo
(
n
)=
k*Vi
(
n
)+(1
−k
)*
Vo
(
n
−1),
with k and 1−k weight factors. The k-factor can be determined by means of a motion detector.
Motion will destroy the confidence in the historic value. Continued temporal filtering in the presence of motion causes motion smear. If a pixel value changes abruptly over time, then it must not be averaged with past values anymore. Protection from smearing can be arranged by using non-linear filtering, i.e., for small temporal differences (k<<1), the present and past values are averaged, and for large differences (k=1), only the present value is used. Temporal averaging removes the temporal high frequencies of the noise, and, thus, it improves the signal-to-noise ratio. It also increases the temporal correlation.
Dynamic noise reduction (DNR) increases the temporal correlation in the picture by temporal low-pass filtering. On a moving picture, this will cause smearing. This is, of course, not desired, so, in the presence of motion, the noise reduction must be switched off. Two kinds of temporal variations must be distinguished, i.e., noise and motion. The threshold may be set according to the current noise level. This is then called “adaptive DNR”.
The disadvantage of such a temporal noise reduction filter is that the k-factor can spatially vary strongly. Further, such a filter requires a relatively large field memory and is therefore rather expensive, while peak noise breakthrough can be a disturbing factor.
SUMMARY OF THE INVENTION
It is, inter alia, an object of the invention to provide an improved noise reduction. To this end, the invention provides a video noise reduction circuit comprising a temporal noise filter, a down-sample unit for obtaining a spatial down-sampling of video signals of subsequent pixels, said down-sampled video signals being supplied to the temporal noise filter, an up-sample unit to regenerate, in response to noise output signals obtained in said temporal noise filter, the noise signals of said pixels, and subtracting means for subtracting said noise signals from the respective video signals of the video signal supplying means.
The consequence of this measure is that not only the k-factor is spatially smoothed, but that also the quantity of information in the temporal noise filter is reduced. This, however, does not lead to any major disadvantage, as in the temporal noise filter, only the obtained spatially low frequency filtered noise signals are supplied to the up-sample unit in order to subtract the, most annoying, low frequency noise from to the video input signals. Further, the number of pixels to be stored in the field memory of the temporal noise filter and to be processed in the temporal noise filter is diminished, which lead to a less expensive and less time consuming processing. As the output of the temporal noise filter is a noise signal rather than a video signal, cheap components may be used for the temporal noise filter, the down-sample unit and the up-sample unit.
To increase the video-image quality, a quincunx down-sampling of video signals of pixels is applied, whereby the odd and even lines within a field have a phase difference corresponding with a down-sampling offset of half of the down-sampling ratio of the video signals. In order to obtain a quincunx down-sampling of video signals, the down-sample unit is preferably provided with two non-recursive discrete transversal filters with mutual asymmetric filter coefficients. In a specific embodiment, filter coefficients (1,1,1,1,0,0)/4 and (0,0,1,1,1,1)/4, respectively, are provided for alternate lines within a field. In this embodiment, video signals of 4 horizontal subsequent pixels are combined in the down-sample unit. However, it will be clear that other filter coefficients and even other filters can be chosen.
In a preferred implementation, the temporal noise filter is formed in such a way that a field memory signal So(m) for a group of down-sampled video signals of subsequent pixels, is a function of the last determined field memory signals So(m−1) for said group of pixels and the down-sampled video input signal Si(m) from the down-sample unit, substantially according to the relation:
So
(
m
)=
Si
(
m
)−(1
−k
)*[
Si
(
m
)−
So
(
m
−1)],
where 1-k is a weight factor, depending on the difference signal dif, formed by the difference b
De Haan Gerard
Schutten Robert Jan
Sijstermans Fransiscus Wilhelmus
Stessen Jeroen Hubert Christoffel Jacobus
Koninklijke Philips Electronics , N.V.
Miller John
Yenke Brian
LandOfFree
Video-apparatus with noise reduction does not yet have a rating. At this time, there are no reviews or comments for this patent.
If you have personal experience with Video-apparatus with noise reduction, we encourage you to share that experience with our LandOfFree.com community. Your opinion is very important and Video-apparatus with noise reduction will most certainly appreciate the feedback.
Profile ID: LFUS-PAI-O-3229589