KR970005833B1 - Adaptive image quality improvement method and apparatus - Google Patents

Abstract

An adaptive picture quality improving device and a method of improving the same are provided to improve the picture quality at a receiving part by effectively removing quantizing noise made by decoding in a conversion region. The adaptive picture quality improving device includes a first predicting means (20) inputting a decoded digital video signal and converting it into a conversion region signal, predicting a power spectrum, and producing a phase component of the digital video; an error modeling means (50) producing a standard deviation relating to a quantizing step used when coding the digital video signal, and a preset random noise with maximum value; a second predicting means (60) inputting the preset random noise and converting it to a conversion region signal, and predicting a power spectrum; an adding means (70) adding the signal power spectrum value from the means (10) and (20) and the noise power spectrum value; and an inversion means (30) inputting the resultant signal power spectrum value from the adding means and the phase component, and converting the resultant digital video signal to convert the resultant signal in the conversion region into a resultant digital video signal in the time zone.

Description

Adaptive image enhancement device and method

1 is a block diagram of an adaptive picture quality improving apparatus according to the present invention.

2 is an explanatory diagram for explaining the operation of the adaptive image quality improving apparatus according to the present invention.

3 is a detailed block diagram of the error modeling apparatus of FIG.

* Explanation of symbols for main parts of the drawings

10: windowing means 20: first predictor

30: inverse converter 40: inverse windowing means

50: error modeling device 60: second predictor

70: adder

The present invention relates to an apparatus for improving adaptive image quality, and more particularly, to an apparatus and a method for improving image quality at a receiver by efficiently removing quantization noise that may occur when encoding a digital image signal.

In the case of transform encoding a digital video signal, an input signal is generally converted into a transform region by a discrete cosine transform and encoded by a quantizer. Since the encoding is performed in the transform domain detail, if a quantization error occurs, there is a deterioration in image quality during the inverse transform in the time domain. However, a method for effectively removing such noise (unusual) has not been identified yet.

On the other hand, in the case of encoding a signal in the time domain, a possible noise may be modeled and removed by modeling a quantization error generated during encoding. The method is as follows.

After converting a video signal with quantization error into the frequency domain, modeling the spectral portion of the error component in the frequency domain detail, subtracting only the error spectral component and converting it into the time domain again, the quantization error known statistically It can be removed approximately. In this case, a method of converting to the frequency domain uses a Discrete Fourier Transform (FFT) and weights a predetermined block as a window function in the time domain to remove the quantization error in units of blocks.

As described above, the existing method of removing the statistical components known statistically can be applied to the coding in the time domain, and there is a limit to the current high definition TV video signal compression method using the transform coding.

That is, since the error in the transform encoding is over the entire frequency domain, if the inverse transform is performed in the time domain again, the error component cannot be easily modeled by the conventional method.

Accordingly, an object of the present invention is to provide an apparatus for improving an image quality that can improve the image quality at a receiver by efficiently removing quantization noise generated during encoding of a transform domain.

Another object of the present invention is to provide an adaptive image quality improvement method capable of adaptively and efficiently removing quantization noise in encoding on a transform domain.

According to an aspect of the present invention, the adaptive image quality improving apparatus of the present invention for removing noise generated when transform encoding a digital image signal is inputted, converts the decoded digital image signal into a transformed region signal, and converts the power spectrum into a transformed signal. First prediction means for predicting and outputting a phase component of the digital image; Error modeling means for outputting a predetermined standard deviation and a predetermined random noise having a maximum value in relation to the quantization step used in encoding the digital video signal; Second prediction means for inputting the predetermined random noise, converting it into a transformed domain signal, and predicting a power spectrum; Adder means for contending the signal power spectral values from the first and second predictor means with a noise power spectral value; And an inverse change means for inputting the resultant signal power spectrum value and the phase component from the adder means to convert the resultant signal on the transform domain into a resultant digital image signal on the time domain.

Hereinafter, described in detail with reference to the accompanying drawings of the present invention.

1 is a block diagram of an adaptive picture quality improvement apparatus according to the present invention, wherein the adaptive picture quality improvement apparatus includes a windowing means 10, a first predictor 20, an inverse transformer 30, a reverse dosing means 40, and an error. And an adder 70 coupling the outputs of the modeling device 50 and the second predictor 60.

Referring to FIG. 2, FIG. 2 shows a two-dimensional image 80 for explaining the operation of the apparatus for improving an adaptive image quality. ) Is a signal containing quantization noise on the transform domain. The signal g (n) is divided into block units of a predetermined size through the windowing means 10, and the weight function W (n) is multiplied and input to the first predictor 20. The video signal g _w (ni, nj) divided by the first predictor 20 is converted into a frequency domain through a Fast Fourier Trnasform (FFT). As shown in FIG. 2, since the input video signal g (n) is two-dimensional, the input video signal g (n) passing through the windowing means 10 may be defined as follows.

Where N and M are integers greater than zero.

When the divided video signal g _w (ni, nj) is converted into a frequency domain through vertical and horizontal discrete Fourier transform, it can be defined as follows.

As shown in FIG. 2, g _w (ni, nj) corresponds to one block of an input video signal corresponding to N horizontally and M vertically, and when the Fourier transform is performed in the horizontal and vertical directions, respectively, The power spectrum of this included receiver is obtained.

In equation (2), the signal G _w ( _ω i, ω j) in the frequency domain is represented by the magnitude component (│G _w ( _ω i, ω j) │) and the phase component θ _w ( _ω i, ω j) for the frequencies _ω i, ω j. The magnitude component │G _w ( _ω i, ω j) │ is supplied to the adder 70 as the power spectral value │G _w ( _ω i, ω j) │ ^{2. The} phase component θ _w ( _ω i, ω j) )) Is input to the inverse transformer 30 to be used for inverse discrete cosine transform.

On the other hand, as will be described in detail later, by the quantization system (θp) from the decoder, the error modeling device 50 is a noise signal (V _w (ni, nj)) of the block similar to the signal block by the above-described windowing Occurs. The noise signal V _w (ni, nj) is converted into the noise signal, and the power spectrum value | V _w ( _ω i, ω j) ² of the noise signal is input to the adder 70.

The adder 70 in the power spectrum value of the aforementioned signal _{(│G w (ωi, ωj)} │ 2) and a power spectrum value of the noise signal there is a coupling _{(│V w (ωi, ωj)} │ 2), the following equation: It may be indicated by.

Here, the value α is a value between 0 and 1, and if the input quantization step θp is large, it is close to 1 because the quantization error is large, and if the input quantization stem θp is small, it is close to zero. Since the above-described compensated signal power value | F _w ( _ω i, ω j) ² must be greater than 0, equation (3) can be redefined by the following equation.

The adder 70 subtracts the signal power specification thromboplastin value includes the quantization error _{(│G w (ωi, ωj)} │ 2) the noise signal power spectrum value from the model _{(│V w (ωi, ωj)} │ 2) from , The compensated signal power spectral value │F _w ( _ω i, ω j) │ ² is input to the inverse transformer 30 and combined with the phase component θ _w ( _ω i, ω j) to inverse discrete Fourier transform, It can be converted into a signal of an area. This can be defined by the equation

In equation (4), IFFT represents the inverse Fourier transform in the horizontal and vertical directions. Here, the compensation signal │F _W (ni, nj) │ is a value that corresponds to the square root of ² │ _w │F (ωi, ωj) obtained in (4). In addition, the phase component θw ( _ω i, ω j) is input from the first predictor 20 as described above. Compensated signal in time domain ( (ni, nj) is output as the signal f (n) compensated by the inverse windowing means 40. It will be appreciated that the resulting compensated signal f (n) is an improved video signal in which quantization noise occurring throughout the transform domain is effectively removed.

In addition, as shown in FIG. 2, since the signal is divided into blocks of a predetermined size and a signal power prediction process, the block in the horizontal and vertical directions is removed to remove the blocking phenomenon occurring at the boundary of the block portion. It is desirable to allow these to overlap each other.

Referring to FIG. 3, a detailed block diagram of the error modeling apparatus 50 of FIG. 1 is shown in FIG. The error modeling apparatus 50 includes a random noise generator 51, a variance and maximum adjustment means 52, an inverse discrete cosine transformer 53 and a windowing means 54.

The random noise generator 51 is a lookup table and has a size of L × L. Where L × L is the magnitude of the signal used for discrete cosine transform during encoding. When the quantization step Qp at the time of encoding is input from the decoding apparatus to the random noise generator 51, the corresponding memorized Gaussian random noise is generated. It can be assumed that there is a random noise of Qp / 2 magnitude in the transform domain. However, since Qp / 2 is the largest error that can occur during quantization, it is necessary to adjust this value appropriately.

The noise signal modeled is output by multiplying the above-mentioned random noise signal γ (ω) so that the standard deviation is Qp / 10.

This signal is again input to an inverse discrete cosine converter 53 and converted into a modeled noise signal V (n) in the time domain. The modeled noise signal V (n) in the time domain is applied to the noise signal _Vw (n) by applying a weight W (n) as in the windowing means 10 described above in the windowing means 24. The noise signal V _w (n) may be predicted as a spectral value in the frequency domain by the Discrete Fourier Transform in the second prediction 60 as described above.

Therefore, the adaptive picture quality improving apparatus of the present invention is coupled between a decoder such as a high-definition TV and a display device, and adaptively models a noise (error) component in relation to the quantization step Qp, and decodes it in the frequency domain. By converting the input image into the time domain in addition to or subtracting from the spectrum of the image, it is possible to effectively and efficiently suppress noise on the transform region generated during transform encoding of a digital image signal.

Although the present invention has been described with reference to representative embodiments, it will be apparent to those skilled in the art that various modifications and variations may be made.

Claims (8)

A first prediction for removing noise generated during transform encoding of a digital video signal, converting the decoded digital video signal into a transform domain signal, predicting a power spectrum, and outputting a phase component of the digital video signal. Means 20; Error modeling means (50) for outputting a predetermined standard deviation and a predetermined random noise having a maximum value with respect to the quantization step (Qp) used for encoding the digital video signal; Second predicting means (60) for inputting said predetermined random noise, converting it into a transformed domain signal and predicting a power spectrum; Adder means (70) for combining the signal power spectral value and the noise power spectral value from the first and second predictor means; And an inverse conversion means (30) for inputting the signal power spectral value and the phase component of the result from the adder means to convert the result signal on the conversion area into a digital image signal of the result on the time domain. The apparatus of claim 1, wherein the image quality improving apparatus inputs the decoded digital image signal, divides the digital image signal into a predetermined size signal block, and multiplies a predetermined weight for each of the divided digital image signals. A windowing means (1) for outputting to the first predictor means (20); And an inverse windowing means (40) coupled to the inverse transform means (30) to restore the blocked digital image in the time domain to its original size, wherein the error modeling means (50) receives the predetermined random noise. And windowing means (54) for dividing the digital video signal into the same shape as the predetermined size signal block. 3. The apparatus of claim 2, wherein the error modeling means (50) comprises: random noise generating means (51) for storing random noise data according to the quantization step; A variance value and a maximum value adjusting means 23 for inputting random noise from the random noise generating means 51 to adjust the standard deviation and maximum value of random noise according to the quantization step, and the adjusted random noise time domain signal. And an inverse discrete cosine converter (23) for converting to the inverse, wherein the windowing means (56) is coupled to the inverse discrete cosine converter (23). 4. The apparatus of claim 3, wherein the standard deviation of the random noise is adjusted to 1/10 of the quantization step and the maximum value of the random noise is adjusted to 1/2 of the quantization step. To remove noise generated when converting and encoding a digital video signal, the decoded digital video signal is input, converted into a transform domain signal, a power spectrum is predicted, and a signal power prediction is output for a phase component of the digital video signal. Step; An error modeling step of outputting a predetermined standard deviation and a predetermined random noise having a maximum value with respect to the quantization step (Qp) used for encoding the digital video signal; A noise power prediction step of inputting the predetermined random noise, converting the signal into a transformed domain signal, and predicting a power spectrum; An addition step of combining the signal power spectral value and the noise power spectral value; And an inverse transforming step of converting the resultant signal on the transform domain into a resultant digital image signal in the time domain by inputting the signal power spectral value and the phase component of the result from the adding step. The method of claim 5, wherein the signal power prediction step inputs the decoded digital video signal, divides the digital video signal into a predetermined size signal block, and multiplies a predetermined weight for each of the divided digital video signals. And a windowing step, wherein the inverse transform step further includes an inverse windowing step of restoring the blocked digital image in the time domain to its original size, and wherein the error modeling step is further configured to convert the predetermined random noise into the digital image signal. And a windowing step of dividing the signal into the same shape as that of the predetermined size signal block. 7. The method of claim 6, wherein the error modeling step comprises: a random noise generation step of storing random noise data according to the quantization step; A variance value and a maximum value adjustment step of adjusting the standard deviation and the maximum value of the random noise according to the quantization step by inputting the random noise, and an inverse discrete cosine transform step of converting the adjusted random noise into a time domain signal And the windowing step is performed following the inverse discrete cosine transforming step. 8. The method of claim 7, wherein the standard deviation of the random noise is adjusted to 1/10 of the quantization step and the maximum value of phase random noise is adjusted to 1/2 of the quantization step.