JPH03155280A - Method for improving image quality due to aliasing distortion - Google Patents

Abstract

PURPOSE:To improve folded distortion by forecasting the generation of folded distortion in each block. CONSTITUTION:An AC component storing memory 32 stores only an AC component in a signal for one block from a 2D-DCT part 31. At the time of ending the conversion of the part 31, an AC component energy calculating part 33 calculates the sum of energy in the AC component for one block by using the information of the memory 32. When the energy sum of the AC component calculated by the part 33 exceeds a prescribed value, a folded distortion forecasting part 34 generates 1-bit informatin in each block. A forecasted result storing part 28 stores information obtained by an AC component forecast inspecting part 27 correspondingly to a block position. An image processing part 29 executes partial image processing in each block based upon the information stored in the part 28. In a part where the generation of folded distortion is forecasted, the signal level of a picture element generating a pattern whose property is sharply different from that of peripheral picture elements due to folded distortion can be corrected based upon relaton with the peripheral picture elements.

Description

[Detailed Description of the Invention] [Summary] Regarding an image quality improvement method in an image encoding system, the present invention provides an algorithm for improving aliasing distortion by processing only on the receiving side, and encoding/decoding on the transmitting side and the receiving side. The purpose is to provide a block distortion improvement method that does not require uniformity of image information, and on the transmitting side, a two-dimensional discrete cosine transform (DC
The DCT transform coefficient sequence obtained by performing T) is encoded and transmitted, and on the receiving side, the O obtained by decoding it is
On the receiving side of a DCT-transform coding system that performs inverse discrete cosine transform on CT transform coefficients to restore image information, the generation of aliasing distortion for each block is predicted from the DCT transform coefficients obtained for each received block. The present invention is constructed by comprising an aliasing distortion occurrence prediction unit that generates an output of a prediction result, and an image processing unit that performs image processing to improve aliasing distortion for the block when the prediction result is output.

[Industrial application field]

The present invention relates to an image quality improvement method in an image encoding system, and particularly relates to a method for improving received image quality deterioration due to aliasing distortion in an image encoding system using two-dimensional discrete cosine transform (DCT). .

In a transform coding system using two-dimensional OCT, on the transmitting side, the original image is divided into blocks of a certain size, two-dimensional DCT is performed on each block, and the obtained transform coefficients are encoded. By doing so, the amount of information can be significantly reduced. On the receiving side, the encoded transform coefficients are decoded and the decoded result is subjected to OCT inverse transform to restore the original image.

In this case, in general, in OCT operation, in order to perform inverse transformation from the transform coefficients obtained by performing DCT forward transform and completely restore the original information, transform coefficients up to an infinite number of terms are required.

However, it is practically impossible to handle coefficients with an infinite number of terms, and processing is usually performed based on approximate expressions that express up to a finite number of coefficients equal to the number of pixels forming a block. For example, if one block is made up of 8x8 pixels,
Coefficients up to the 64th term are used.

For this reason, especially in blocks where the dispersion of the signal levels of pixels within the block is large, that is, blocks with many high frequency components, the transform coefficients obtained as a result of the DCT forward transform may be It is not possible to completely restore the information, and patterns of information that do not exist in the original information may occur. Image distortion due to such causes is called aliasing distortion.

When aliasing distortion occurs, information with a pattern that does not exist in the original information appears, which significantly deteriorates the impression and quality of the received image.

Therefore, in order to construct a practical coding system with high quality and high efficiency, it is necessary to prevent aliasing distortion from occurring even in the DCT operation of finite terms, or to prevent aliasing distortion from occurring even if aliasing distortion occurs. There is a need for an image quality deterioration improvement method that makes image quality deterioration visually inconspicuous (hard to see).

[Conventional technology]

Figures 6(a) and 6(b) show the basic configuration of a DCT transform encoding system, illustrating one that performs layered encoding processing, and (a) shows the transmitting (encoding) side. shows,
) indicates the receiving (decoding) side.

In FIG. 6(a), a luminance signal Y1 and a color difference signal U.

The image information consisting of 1 is stored in the frame memory 11.
After the screen is accumulated, low pass filter (LPF) 1
In a preprocessing section consisting of 2 and a subsampling section 13, preprocessing is performed for each block consisting of, for example, 8x8 pixels. That is, the image information in the frame memory 11 is band-limited in the LPF 12, and then sampled in the vertical (V) direction and horizontal (f() direction at a ratio of 2:1 in the sub-sampling unit 13, respectively, to compress the amount of information. This image information is again input to the LPF 12 and band-limited, and then sampled at a ratio of 2:1 in the V direction and the H direction in the sub-sampling section 13 to compress the amount of information.Such processing is repeated. For example, after the amount of information is compressed to 1/16,
In the subtracter 14, subtraction is performed block by block between the predicted picture and the information of the difference picture.

The information of the difference image is processed by two-dimensional discrete cosine transformation (2D-D
In the CT) unit 15, a two-dimensional discrete cosine transform is performed for each 8×8 pixel block, converting a time domain signal into a frequency domain signal, and generating a series of transform coefficients representing information of the difference image. .

The variable length entropy encoding unit 16 generates the shortest code word for the information of this transform coefficient sequence.

A coded signal consisting of a variable length code is generated by performing coding that allocates a variable length code. This encoded signal is 1 as described above.
This is information of a differential image whose information amount has been compressed to /16.

On the other hand, the two-dimensional discrete cosine inverse transform m (2D-IDCT) unit 17 performs two-dimensional discrete cosine inverse transform on the information of the transform coefficient sequence of the 2D-DCT unit 15, so that
Restore the difference image. The adder 18 adds the difference image information and the predicted image information block by block to generate restored image information. The linear interpolation unit 19 performs linear interpolation on the information of the restored image, and generates image information with twice the amount of information in each of the V direction and the H direction. The frame memory 20 stores the image information of the linear interpolation unit 19 for one screen, thereby generating predicted screen information whose information amount is compressed to 1/4.

Next, in the preprocessing section, the same processing as the previous time is performed again on the image information whose information amount has been compressed to 1/4, thereby generating an encoded signal for the information of the difference image whose information amount has been compressed to 1/4. At the same time, linear interpolation is performed on the information of the restored image restored from this to generate predicted screen information having the same amount of information as the original image.

Furthermore, in the preprocessing section, the same processing is performed again on the image information whose information amount is not compressed, thereby generating an encoded signal for the information of the difference image with respect to the original image.

In FIG. 6(b), the variable length code decoding unit 21 decodes the variable length code sent from the transmitting side and reproduces a series of transform coefficients. The 2D-IDC 7 unit 22 restores a difference image by performing two-dimensional inverse discrete cosine transform on the information of this transform coefficient series. The adder 23 adds the difference image information and the predicted image information block by block to generate restored image information, and stores the information in the frame memory 2.
4 restores the original image by accumulating this image information for one screen. The programmable linear interpolation unit 25 performs linear interpolation on the information of the restored image at a rate that changes depending on the compression ratio, and generates an image output signal.

On the other hand, the linear interpolation unit 26 performs linear interpolation on the restored image in the frame memory 24 to generate image information with twice the amount of information in the V direction and in the H direction. This image information is used as the above-mentioned predicted screen information.

FIG. 7 is a diagram illustrating the basic operation flow of the DCT transform encoding system.

■ Divide the original image into blocks (subpictures) of a certain size.

■ Next, two-dimensional discrete cosine transformation is performed for each block.The two-dimensional discrete cosine transformation is expressed by the following formula%, where X (i, j
) is the original image block, A is a constant, Y (i, Jl
is the transform coefficient array.

(2) Image information is reconstructed using the transform coefficients obtained for each block in this way.

■Next, select the conversion coefficients to be transmitted.

This is done by thresholding, subsampling of transform coefficients, etc.

■Next, quantize the transformation coefficients taking visual characteristics into consideration.

(2) Code words are assigned to the transform coefficients quantized in this way.

When encoding is performed in a hierarchical manner in this way, the first encoded signal with the compressed amount of information has a small amount of information, so the screen definition is low, but the encoding processing time is short, so when switching screens , one screen of information can be sent quickly.

The amount of information in this image increases each time it is encoded, so the definition gradually increases, and eventually the information of the original screen is sent.

Therefore, according to such a layered encoding method, when still images are intermittently switched and transmitted, it is possible to prevent blanks from appearing on the screen because it takes time to transmit the first image when switching the screen.

In this way, when image information is encoded and transmitted block by block and decoded on the receiving side to restore the original image,
Since the transform coefficients used in the DCT forward transform and inverse transform have a finite number of terms, the reproduced information is not completely restored, and aliasing distortion may occur and image quality may deteriorate.

Conventionally, various methods such as those described below have been considered as methods for improving image quality deterioration caused by such aliasing distortion.

(2) The size of one block is made sufficiently small in order to make the dispersion of signal levels of the pixels forming the block sufficiently small (high frequency components are small).

■ Conversion coefficients up to the number of terms at which aliasing does not occur are treated as effective coefficients.

■ By increasing the calculation precision of the conversion coefficients to a very high level, the effects of using approximate expressions for the conversion coefficients are reduced.

[Problem to be solved by the invention]

The conventional methods for improving aliasing distortion as described in the previous section each have the following problems.

(1) In the method of reducing the size of one block shown in item (2) in the previous section, the number of blocks constituting an image increases, so the time required for processing increases. In addition, the area of the image that can be expressed by one DCT operation is small (as a result, the efficiency of compressing the amount of information by encoding decreases).

(2) The method of treating the conversion coefficients up to a large number of terms as effective coefficients as shown in (2) in the previous section not only complicates calculations but also causes an increase in the amount of information in the generated code.

(3) The method of increasing the calculation accuracy of transform coefficients shown in item (■) in the previous section not only makes DCT calculations complicated, but also leads to an increase in the hardware scale because it handles transform coefficients with very high precision. .

(4) In any of the methods described in the previous section, in order to implement countermeasures, a coding system based on the same idea that assumes that both the transmitting side and the receiving side are compliant with the deterioration improvement method must be configured. This reduces the versatility of the encoding system.

The present invention is an attempt to solve the problems of the prior art, and is capable of improving aliasing distortion only by processing on the receiving side, and by using encoding/decoding algorithms on the transmitting and receiving sides. The purpose of this invention is to provide a method for improving image quality deterioration due to aliasing distortion, which does not require uniformity.

That is, the present invention attempts to solve the problems of the prior art as follows.

■ Even if you configure the coding system by setting the block size and DCT variation accuracy to practical levels (
In other words, even if some aliasing distortion is allowed),
By introducing an image processing method on the receiving side, the resulting aliasing distortion becomes visually invisible.

■ By making it possible to take measures to improve distortion through processing only on the receiving side, flexibility is left in the processing on the transmitting side (encoding side) and the encoded data structure, making it possible to configure a general-purpose encoding system. enable.

[Means to solve the problem]

As shown in FIG. 1, the basic configuration of the present invention is such that, on the transmitting side, two blocks of image information are
The DCT transform coefficient series obtained by performing dimensional discrete cosine transform (hereinafter abbreviated as DCT) is encoded and transmitted, and on the receiving side 2, the discrete cosine inverse is applied to the DCT transform coefficients obtained by decoding this. By providing an aliasing distortion generation prediction unit 3 and an image processing unit 4 on the receiving side 2 of the DCT transform coding system that performs transformation to restore image information, adaptive aliasing can be generated by processing only on the receiving side. This is intended to improve image quality deterioration.

Here, the aliasing distortion occurrence prediction unit 3 predicts the occurrence of aliasing distortion for each block from the DCT transform coefficients obtained for each received block, and generates an output of the prediction result, and the image processing unit 4 outputs the prediction result. When this output occurs, image processing is performed on that block to improve aliasing distortion.

[Effect]

On the transmitting side of the DCT transform coding system, the difference between the image information and the predicted image is calculated for each fixed block of image information. Then, a two-dimensional discrete cosine transform (DCT) is performed on this difference, and the resulting DCT transform coefficient sequence is encoded and transmitted to the receiving side. On the receiving side, the DCT transform coefficients obtained for each block are decoded and subjected to inverse discrete cosine transform to reproduce a difference image, which is then accumulated to restore the original image information.

In the present invention, on the receiving side, it is predicted from the DCT transform coefficients of each transmitted block whether or not aliasing distortion will occur in that block. This prediction is, for example, A in the DCT transform coefficients in that block.
This can be done by testing the C component.

When the occurrence of aliasing distortion is predicted, image processing is performed to improve the aliasing distortion for that portion. Image processing in this case is, for example, "noise removal while preserving edges."
This can be done by applying the following method.

In this way, the receiving side estimates the location where aliasing distortion occurs and performs image processing on that area to improve image quality deterioration caused by aliasing distortion. It becomes possible to improve the deterioration.

〔Example〕

FIG. 2(a), Φ) shows the basic configuration of the present invention, and the sixth
An example is shown that performs layered encoding processing in the same way as in Figures (a) and (b), where (a) shows the sending (encoding) side and (b) shows the receiving (decoding) side. Figure 6 (a)
).

The same items as in (c) are indicated by the same numbers, and 27 is A.
The CC component seed type testing unit 28 is a prediction result storage unit, and 29 is an image processing unit.

In the present invention, the aliasing distortion removal process is performed only on the receiving (decoding) side, and therefore the second
The configuration of the transmission (encoding) side shown in Figure (a) is shown in Figure 6 (a).
This is exactly the same as the conventional case shown in .

On the receiving side shown in FIG. 2, etc., the variable length code decoding unit 21 decodes the variable length code to reproduce the transform coefficient sequence, and the 2D-IDC 7 unit 22 performs two-dimensional discrete cosine inverse transform on the information of the transform coefficient sequence. generation of restored image information by adding the difference image information and predicted screen information in the adder 23, and restoration of the original image by accumulating one screen worth of restored image information in the frame memory 24. , generation of an image output signal by linear interpolation at a rate that changes according to the compression ratio for the information of the restored image in the programmable linear interpolation unit 25; generation of predicted screen information by linear interpolation of the restored image in the frame memory 24 in the linear interpolation unit 26; The occurrence is shown in Figure 6 (c)
This is done in the same way as in the case of

The AC component measurement test unit 27 predicts the location where aliasing distortion occurs by testing the magnitude of the AC component coefficient among the DCT transform coefficients of the target block in the received image. The prediction result is output as a signal expressing the possibility of occurrence of aliasing distortion using 1-bit information.

The prediction result storage unit 28 constitutes a frame memory for the number of blocks in one screen, and stores 1-bit information indicating the prediction result of the aliasing distortion occurrence point in the AC component measurement verification unit 27, corresponding to the position of each block. memorize it.

The image processing unit 29 refers to the prediction results for each block stored in the prediction result storage unit 28, performs processing to remove aliasing distortion on the portion where aliasing distortion is expected to occur, and then performs processing to remove the aliasing distortion. Create image information to be displayed.

FIG. 3 shows an embodiment of the present invention, showing only the receiving (decoding) side, where the same parts as in FIG. It is. In addition, in the AC component measurement verification section 27, 31 is 2D-
A DCT section, 32 is an AC component storage memory, 33 is an AC component energy calculation section, and 34 is an aliasing distortion prediction section.

In the embodiment shown in FIG. 3, the display memory 30 stores the linearly interpolated image output signal from the programmable linear interpolator 25 for one screen. The image signal for one screen stored in the display memory 30 is used for display on a display or the like (not shown). Furthermore, the AC component submeasurement testing unit 27 calculates the energy of all AC components of the target block with respect to the AC component coefficients in the received transform coefficients.

In the AC component measurement verification section 27, the 2D-DCT section 31
performs a two-dimensional discrete cosine transform on each block of the 100 restored images stored in the frame memory 24, and generates a transform coefficient series representing the information of the difference image, which is composed of frequency domain signals. occurs. The AC component storage memory 32 stores only the AC component that is being output from the 2D-DCT section 31 in the frequency domain for one block. A.C.
When the 2D-DCT section 31 completes the conversion, the component energy calculation section 33 uses the information in the AC component storage memory 32 to calculate the total energy of the AC components for one block. When the total energy of the AC components calculated by the AC component energy calculation unit 33 exceeds a predetermined value, the aliasing distortion prediction unit 34 recognizes the block as having a large value up to the high-order transformation coefficients, and calculates the aliasing distortion. One bit of information is generated for each block to determine where the occurrence is likely to occur.

The prediction result storage unit 2 stores the region division information in units of blocks obtained by the AC component measurement verification unit 27 in correspondence with block positions.

The image processing unit 29 performs partial image processing on a block-by-block basis based on the area division information stored in the prediction result storage unit 28. At this time, by not performing any processing on areas where aliasing distortion is not expected to occur, the image quality is reduced due to meaningless image processing on areas unrelated to aliasing distortion that occurs when processing to improve image quality deterioration. Prevent decline.

For areas where aliasing distortion is expected to occur, for example, ``edge-preserving noise removal'' is a known image processing method that removes partial noise while preserving the original information structure. By applying this method, it is possible to correct the signal level of a pixel in which a pattern that is significantly different from the properties of the surrounding image due to aliasing distortion is generated based on the relationship with the surrounding pixels.

FIG. 4 shows an example of a test regarding AC component coefficients of DCT transformation coefficients.

The target block in the received image consists of image information for one block consisting of MXN pixels (M=N-4 in FIG. 4). In contrast, the two-dimensional discrete cosine transform (
2D-DCT) to generate transform coefficients consisting of DC and AC components. Among these, the AC component coefficient energy is calculated. The total energy E of the AC component is calculated by the following equation.

In the second row, the y-th value is the value of the conversion coefficient (signal level), y. is the DC coefficient value (signal level), and M and N are the block sizes.

Next, the occurrence of aliasing distortion is predicted depending on whether E>T)I, where E is the total energy of the AC component,
TH is a predefined aliasing distortion occurrence prediction coefficient, and is defined for the total energy 31E.

When the judgment result E>TH, the prediction result is “1”.
'' is written in the prediction result storage unit 28. If not, “0” is written in the prediction result storage unit 28 as the prediction result.

Such processing is sequentially performed for each block, and when it has been performed for all blocks, the processing is terminated.

FIGS. 5(a) and 5(b) illustrate nine noise removal techniques that preserve edges (smoothing that preserves edges).

Now, assuming that the target pixels are X and Y, refer to the surrounding 25 pixels and form a pentagon or six pixels as shown in Figure 5(a).
Eight types of masks including square target pixels and Figure 5(b)
By setting the average value of the mask with the least variance as the value of the target pixel among a total of nine types of masks including 3 × 3 pixel masks surrounding the target pixel as shown in Figure 2, the noise is removed while preserving edges. can be removed.

This method can also be used as a simple method for sharpening blurred edges.

〔Effect of the invention〕

As explained above, according to the present invention, in the two-dimensional DCT transform encoding method, the receiving side predicts the location where aliasing distortion occurs and performs image processing on that part to improve the aliasing distortion. Therefore, even for a variety of information such as natural images where the generation pattern cannot be predicted by processing on the receiving side alone, it is possible to respond to the partial nature of the image.
It is possible to implement processing to improve image quality deterioration due to aliasing distortion.

According to the present invention, in a two-dimensional DCT transform coding system, sufficient effects can be expected as long as basic elements such as block size and representation accuracy of transform coefficients are specified, so that high performance can be achieved with simple processing. quality.

It becomes possible to realize a highly efficient encoding system.

[Brief explanation of the drawing]

Fig. 1 is a diagram showing the basic configuration of the present invention, Fig. 2(a),
Φ) is a diagram showing the basic configuration of the present invention, FIG. 3 is a diagram showing an embodiment of the present invention, FIG. 4 is a diagram showing an example of verification regarding AC component coefficients of DCT conversion coefficients, and FIG. a), (
b) is a diagram explaining the noise removal method that preserves edges, Figures 6(a) and (b) are diagrams showing the basic configuration of the DCT transform encoding system, and Figure 7 is a diagram showing the basic configuration of the DCT transform encoding system. It is a diagram explaining the flow of basic operation. 1.2 indicates a DCT transform coding system, l is a transmitting section, 2 is a receiving section, 3 is an aliasing distortion occurrence prediction section, and 4 is an image processing section.

Claims (1)

[Claims] A DCT transform coefficient sequence obtained by performing two-dimensional discrete cosine transform (hereinafter abbreviated as DCT) on image information for each predetermined block on the transmitting side (1) is encoded and transmitted. , At the receiving side (2) of the DCT transform encoding system, the DCT transform coefficients obtained by decoding the DCT transform coefficients are subjected to inverse discrete cosine transform to restore image information. an aliasing distortion occurrence prediction unit (3) that predicts the occurrence of aliasing distortion for each block from the DCT transform coefficients obtained for each block and generates an output of a prediction result, and improves the aliasing distortion for the block when the prediction result is output. An image quality deterioration improvement method due to aliasing distortion, characterized by comprising an image processing section (4) that performs image processing.