Image analysis – Image enhancement or restoration – Artifact removal or suppression
Reexamination Certificate
1998-12-29
2002-06-18
Lee, Thomas D. (Department: 2724)
Image analysis
Image enhancement or restoration
Artifact removal or suppression
C382S274000
Reexamination Certificate
active
06408104
ABSTRACT:
BACKGROUND OF THE INVENTION
1. Field of the Invention
The present invention relates to elimination of noise in a video signal encoder and more particularly to a method and apparatus for eliminating noise in a video signal encoder, which computes energy of a noise component and a pure signal component and controls a variable gain over a filtering path based upon energy content of the noise component with respect to whole signal components.
2. Description of Related Art
Noise elimination performed in a conventional video signal encoder, for example, a moving picture experts group (MPEG)2 video signal encoder, depends on gain control which is represented with “1−a” as shown in
FIG. 1
over a path of a signal.
FIG. 1
shows a video signal encoder employing a conventional noise elimination method. The encoder comprises: subtracter
101
for extracting an error signal from an input video signal which is processed in the unit of blocks; gain variable part
102
for performing a gain control operation of “1−a” (“a” is a gain) with respect to an output signal of the subtracter
101
; gain control part
103
for controlling a gain of the gain variable part
102
; discrete cosine transform (DCT)/quantization (Q) part
104
for performing DCT and quantization with respect to an output signal of the gain variable part
102
; inverse quantization (IQ)/inverse DCT (IDCT) part
105
for performing inverse quantization and inverse DCT with respect to the discrete cosine transformed and quantized signal; adder
106
for summing the inverse discrete cosine transformed signal and a motion compensated signal; memory
107
for storing the summed signal; and motion estimation and compensation part
108
for performing motion estimation and motion compensation using an output signal of the memory
107
and an input video signal.
Referring to
FIG. 1
having such configuration, an input signal, g(i,j,k), (“i,j” is a two dimensional coordinate of a video processing unit on a frame and “k” is time domain information corresponding to the “i,j” in a time frame) is a combination of noise (n(i,j,k)) and pure signal (f(i,j,k)).
Accordingly, a pixel, f(i−dx,j−dy,k−1), in a previous frame is subtracted from a current frame pixel, g(i,j,k), so as to calculate a difference between the current frame pixel and the previous frame pixel. A gain value, a, is controlled based upon the difference. Noise is eliminated in such a manner of reducing the noise by increasing the gain, a, when the difference between the current frame pixel and the previous frame pixel is smaller than a predefined threshold.
Specifically, once receiving the input signal, g(i,j,k), the subtracter
101
subtracts a pixel, f(i−dx,j−dx,k−1), corresponding to the input signal, g(i,j,k), (the pixel has been predicted by the motion estimation and compensation part
108
) from the input signal, and the subtractor
101
outputs a result of the subtraction to respective the gain variable part
102
and the gain control part
103
. The gain control part
103
increases or decreases the gain, a, of the gain variable part
102
according to the output of the subtractor
101
, that is, the scale of the difference between the two frames, thus reducing the noise.
For example, as shown in
FIG. 2
, the subtracter
101
calculates a difference between an input signal and a predicted signal as follows: d(i,j,k)=g(i,j,k)−f(i−dx,j−dx, k−1)=f(i,j,k)−f(i−dx,j−dx,k−1)+n(i,j,k). If the motion estimation and compensation is exact, the difference signal, d(i,j,k), is mainly composed of noise, n(i,j,k), so the gain control part
103
adjusts the gain, a, of the gain variable part
102
to on the order of ‘1’, thus eliminating all of the d(i,j,k). On the other hand, if the motion estimation and compensation is not exact, the d(i,j,k) may be detected to be mainly composed of a signal component, f(i,j,k)−f(i−dx,j−dx,k−1), so the gain control part
103
adjusts the gain, a, to on the order of ‘0’, thus not eliminating the d(i,j,k) even though it may not eliminate the noise component, n(i,j,k).
In other words, as illustrated in
FIG. 2
, the gain control part
103
performs an absolute operation with respect to the difference signal between the input signal and the motion compensated signal, and compares an average of the result values to a reference value, thus controlling increase and decrease of the gain, a, of the gain variable part
102
according to the scale of a difference between the two values.
The gain control is performed only in an INTER mode. In an INTRA mode, the gain control process is bypassed. For this operation, a switching part which switches according to the INTER/INTRA mode is coupled to a front end of the DCT/Q part
104
.
In the INTRA mode, a macroblock is encoded using only information on its own macroblock, so the motion estimation is not performed. On the other hand, in the INTER mode, the macroblock is encoded using information on another macroblock at another time as well as its own macroblock, so the motion estimation is performed.
A signal where the gain, a, has been controlled and noise has been reduced is input to the DCT/Q part
104
. The signal is converted into frequency information through DCT and quantization by the DCT/Q part
104
. The quantized signal is sent to output part
109
for transmission and, simultaneously, sent to the IQ/IDCT part
105
. The IQ/IDCT part
105
performs dequantization and IDCT of the quantized signal and then sends a result signal to the adder
106
. The adder adds the inverse discrete transformed signal to the motion compensated signal from the motion estimation and compensation part
108
so as to reconstruct an original signal. The reconstructed original signal is stored in the memory
107
.
The signal stored in the memory
107
is input to the motion estimation and compensation part
108
. The motion estimation and compensation part
108
performs motion estimation and compensation with respect to a current frame based upon the signal stored in the memory
107
and then sends the result to the respective subtracter
101
and adder
106
.
The output part
109
performs variable length coding (VLC) and first input first output (FIFO) with respect to the noise eliminated and quantized signal and then the result signal is output in the form of a bit stream.
Such video signal encoder to which the conventional noise elimination method is applied uses a scale of the sum of the difference signals as a reference for determining how many signal components, f(i,j,k)−f(i−dx,j−dx,k−1), are contained in the difference signal, d(i,j,k).
Since the signal components and noise components are all together contained in the sum of the difference signals without discrimination, it cannot be identified which of the signal and noise is larger, so the system cannot properly manage the case that the noise components are much more than the signal components or the opposite case. Additionally, unnecessary filtering, such as elimination of a video signal as well as the noise, may be performed, decreasing efficiency in the noise elimination.
The conventional video signal encoder performs noise filtering in the INTER mode, so there occurs problem that the noise does not eliminated over a path where the filtering is not performed. Since the conventional video signal encoder uses only one previous pixel, the noise is not completely eliminated, and, moreover, blocking effect may increase.
FIGS.
3
(
a
) to
3
(
d
) illustrate the blocking effect.
Since the motion estimation and compensation is performed with respect to original image in the unit of blocks, as shown in FIG.
3
(
a
), discontinuities
201
may occur at block boundaries in predicted image.
Since the gain control (1−a) for mc error image(the error signal is indicative of a difference between an input signal including noise and a predicted signal as shown in FIG.
3
(
b
)), distorts block boundary of the mc e
Cho Sang-Hee
Lee Hee Sub
Lim Il-Taek
Lim Kyoung-Won
Min Cheol-Hong
Brinich Stephen
Fleshner & Kim LLP
Lee Thomas D.
LG Electronics Inc.
LandOfFree
Method and apparatus for eliminating noise in a video signal... does not yet have a rating. At this time, there are no reviews or comments for this patent.
If you have personal experience with Method and apparatus for eliminating noise in a video signal..., we encourage you to share that experience with our LandOfFree.com community. Your opinion is very important and Method and apparatus for eliminating noise in a video signal... will most certainly appreciate the feedback.
Profile ID: LFUS-PAI-O-2913459