JPS62104283A - Noise reduction system for differential decoding signal in animation picture transmission - Google Patents

Abstract

PURPOSE:To reduce the noise component remarkable at a flat portion without giving fog to the edge by providing a noise reduction processing forecasting the position of the edge part and acting like portions only except a forecast position selectively to an inverse orthogonal conversion output. CONSTITUTION:The noise reduction processing section 7 shown in broken lines is added to the constitution of the orthogonal conversion coding system. A mask forming section 8 reads the content of portions corresponding to an existing input block position from a difference memory section 10 to form the same size of mask as that of the input block. Let the (i, j)-th picture element be d(i, j), then it is discriminated that the d(i, j) is significant when ¦d(i, j)¦>=Th, where Th is a predetermined threshold value and m(i, j)=1 is established. The magnification processing of 1-picture element region is applied to the obtained mask to give a margin to the forecast position of the edge. A mask processing section 9 applies processing for noise reduction selectively to the inverse orthogonal conversion output based on the mask obtained by the mask forming section 8.

Description

Detailed Description of the Invention (Technical Field of the Invention) The present invention relates to a local differential decoding signal on a transmitting side in a moving image transmission system that transmits a differential signal that changes over time between frames of a moving image signal to be transmitted. Or it relates to noise reduction of differentially decoded signals on the receiving side.

(Background of the Invention) When transmitting a television signal (hereinafter referred to as TV signal), there are two methods: one is to transmit it as waveform information in an analog manner, and the other is to convert it into a digital signal through sampling and quantization operations and then transmit it. There is. Here, we will consider the latter method.

When a normal TV signal is digitized, the amount of information becomes enormous, approximately 90 megabits per second. On the other hand, when considering video conferences, etc., there are
1.5 to 2 megabits per second, or even 384 kilobits per second
It is required to transmit moving image information with a low amount of information on the order of bits.

In order to satisfy this requirement, it is necessary to use 1745 bits per second for the original moving image signal.
~l/60, 1/24 at 384 kilobits per second
The amount of transmitted information must be reduced to 0.

One method for achieving such a high information compression rate is to obtain an interframe difference signal and then perform orthogonal transformation on it. FIG. 1 is a basic configuration diagram of a conventional orthogonal transform encoding method for interframe difference signals. Figure 1 (a
) is a block diagram when motion compensation is not performed, and includes a blocking operation unit 1 for rearranging input image signals to put them in block order, and an image memory unit for storing image signals for at least one frame. 14, a difference detection unit 2 that calculates the difference in pixel units (inter-frame difference or motion compensated inter-frame difference) between the input block and the corresponding block of the image memory unit 14, and a difference detection unit 2 that calculates the difference in pixel units (inter-frame difference or motion compensated inter-frame difference) between the input block and the corresponding block of the image memory unit 14, and An orthogonal transformation unit 3 performs orthogonal transformation, a quantization unit 4 performs quantization on orthogonal transformation results, and the quantized output is branched, one transmits it to the receiving side, and the other transmits the quantization result block by block. an inverse orthogonal transform unit 6 for performing inverse orthogonal transform and local decoding; and an adder 13 for calculating the sum of the inverse orthogonal transform result and the corresponding block of the image memory unit 14 in pixel units. It is composed of.

The image memory unit 14 stores the local decoding result of one frame before, and this becomes the predicted value for the input pixel signal in the current frame. The difference between the input pixel signal and this predicted value for each pixel in a rectangular block, that is,
Find the interframe difference signal. Next, the obtained interframe difference signal is subjected to orthogonal transformation, the transformation result is further quantized, and the quantized output is transmitted to the receiving side.

In addition, in order to obtain local decoding results on the transmitting side, an inverse orthogonal transform is performed on the quantized output, and the image memory unit 14
After adding up the predicted value stored in , the result is stored in the corresponding area in the image memory unit 14 . In the above process, by calculating the inter-frame difference, the redundancy of image data in the temporal direction can be reduced, and by performing orthogonal transformation, the redundancy of image data can be reduced in the spatial direction. As a result, the amount of information regarding one density color can be reduced.

Note that the receiving side also decodes the received quantized signal by performing inverse orthogonal transform, but since this is exactly the same process as the local decoding described above, the following explanation will focus on the receiving side decoding and the transmitting side local decoding. These processes are collectively referred to as "decoding."

Orthogonal transformation is defined as follows. For an image X of size MXN pixels, by the unitary matrix AM, As, X=AM
X is the transformation result and consists of MXN elements) A transformation of the form is called a two-dimensional unitary transformation, and especially when the unitary matrix is an orthogonal matrix, it is called a two-dimensional orthogonal transformation. From the point of view of application to high-efficiency coding, Walsh-Hada
Examples include mard transformation and discrete CO3 transformation (hereinafter also referred to as DCT). As an orthogonal matrix, for example, the size is 4×4
In this case, it becomes like a garment.

FIG. 1 (bl is a block diagram when motion compensation is performed.

In FIG. 1(a), the pixel values in the block located at the same position as the input block are read out from the image memory unit 14, whereas in FIG. A motion detection unit 16 is provided to search for a block most similar to the block, and a motion compensation prediction unit 15 is provided to use a pixel value in the detected block as a predicted value for the pixel value in the input block, that is, a motion compensated predicted value. The difference between the pixel value and the pixel value within the input block, that is, the motion compensated inter-frame difference signal is determined.

The subsequent operations are the same as those in Figure 1 (al).For input images where objects are moving on the screen, motion compensation is performed to select the block most similar to the input block from the image memory. Since a block is searched and a difference between the blocks is determined, it is generally possible to more effectively reduce the redundancy of image data in the temporal direction than when motion compensation is not performed.

In the case of the method shown in FIG. 1, the interframe difference signal (approximately the same applies when motion compensation is performed) is mainly generated in areas where an object moves or brightness changes between frames. For this reason, the difference signal rarely has a distribution in which the value changes smoothly, but rather a linear distribution, depending on the amount of movement between frames.

Large values often occur in bands or dots.

When performing orthogonal transformation on such a signal, the power is not necessarily concentrated in the low-order components, and the high-order components become noticeable in addition to the low-order components, thus increasing the amount of information generated. It turns out. On the other hand, the aforementioned 384 kilobits/
In order to transmit differential signal values at a bit rate as low as a second, there is a requirement that coarse quantization must be performed on orthogonal transform outputs in order to reduce the amount of generated information.

When inverse orthogonal transformation is performed on the transmitting and receiving sides, the output is the sum of products of each order component and a basis function representing the inverse orthogonal transformation. Noise occurs in the decoded signal due to quantization errors in each order component. This will be explained using DCT as an example. D.C.
In the case of T, the basic waveform is a cos function, and for example, for a one-dimensional image consisting of four elements, the first element of the inverse DCT output is expressed as follows.

(However, a0 to a3 are DCT (forward transform) on the input image.
The value when a0 to a3 are quantized is a゛. ˜a”3, the first element in the inverse OCT output at this time is.

In particular, if it becomes a"l""a'3"O after quantization, then the following is true. In other words, it would originally be expressed by only a combination of 5 basis functions.

For this reason, it depends on the size of each order component of 80 to a, and therefore the degree of contribution to the inverse DCT output.
Quantization (especially coarse quantization) causes noise in which the density changes in a wave-like manner.In addition, the way the noise appears is that when the contribution of middle and high-order components is large, Rather than a wavy change in shading, it is observed as a slightly more detailed, rustling change in shading, also known as haze noise.In either case, this type of noise is caused by sharp changes in shading and edges. It is particularly noticeable in flat parts of blocks where there are other flat parts mixed in, and is a major cause of image quality deterioration. It is necessary to take measures to increase or reduce the

(Prior Art and its Problems) Next, a conventional method for removing or reducing this type of noise will be described.

Regarding orthogonal transform encoding, attempts have been made to apply it to original image signals (intra-frame or intra-field encoding). In the case of an original image signal, since there is a high correlation between neighboring pixels as a general property of natural images, when this signal is orthogonally transformed, power tends to be concentrated in low-order components. Furthermore, in the past, there was little demand for the high information compression rate mentioned above, and therefore quantization could be performed finely, and noise such as the above-mentioned wavy changes in shading or buzzing noise could be avoided. It wasn't a problem.

In addition, attempts have been made to orthogonal transform coding for inter-frame difference signals, but in the past, sufficient consideration has not been given to achieving a high information compression rate as mentioned above, and therefore, it has not been possible to effectively reduce the noise as mentioned above. No reduction method has been reported.

On the other hand, as a general method for reducing noise in locally decoded images or decoded images in high-efficiency encoding, the second method is
As shown in the figure, a configuration may be considered in which a low-pass filter section 17 is inserted in the path for storing the decoded image in the image memory section, but since it is applied to the entire screen, the noise component is reduced and at the same time There is a problem that blurring is observed. As the low-pass filter section 17, instead of using a simple filter, for example, considering a 3×3 pixel range as shown in FIG. 3, a filter that performs the following operations for the central pixel E can also be used. .

+ B-HI + I D-F When I≧Th, it is determined that it is an edge portion, and the value of E is output as is.

IB-HI+1D-Fl<Th When , the low-pass filter processing result is substituted into E.

(However, Th is an appropriate threshold value.) Through such an operation, low-pass filter processing can be selectively performed. However, if the ThO value is lowered, there is a risk that the noise component will be judged as an edge part and remain.
Conversely, if the value of Th is increased, the original edges in the image will become blurred.

As described above, in orthogonal transform for inter-frame difference signals or motion-compensated inter-frame difference signals, the conventional technology has no problem with regard to the above-mentioned noise components when coarse quantization is introduced to achieve a low bit rate. No effective reduction method has been found.

Furthermore, similar noise components occur during decoding on the receiving side, leading to deterioration in the quality of the decoded image signal.

(Objects and Features of the Invention) The present invention has been devised in view of the above-mentioned drawbacks of the prior art. The purpose of this paper is to realize a noise reduction method for differentially decoded signals in moving image transmission that enables quality improvement.

The feature of the present invention is to predict the position of a region where the inter-frame difference signal or motion-compensated inter-frame difference signal is likely to become large, that is, the edge part, in the current frame, and to calculate the part excluding the predicted position from the inverse orthogonal transform output. The reason is that a noise reduction processing section is provided that selectively performs noise reduction processing only on the noise.

(Structure of the invention) The present invention will be described in detail below using the drawings.

Note that in the following explanation, a noise reduction method on the transmitting side will be explained in detail by taking as an example.

FIG. 4 shows an embodiment of the present invention, in which the interframe difference signal (
FIG. 2 is a block diagram of an orthogonal transform encoding method for motion-compensated inter-frame difference signals.

As is clear from the figure, the present invention adds a noise reduction processing section 7 surrounded by a broken line to the configuration of the conventional orthogonal transform encoding system. This noise reduction processing section 7 is a mask creation section 8.
．． Mask processing section 9. Difference memory section 10 and branch section 11
It is composed of. In the following description, the description of parts that overlap with those in FIG. 1 will be omitted, and only the parts constituting the noise reduction processing section 7, which is a feature of the present invention, will be described in detail. The difference memory unit 10 converts the output of the inverse orthogonal transform unit 6 corresponding to the decoded value of the inter-frame difference signal into a small number (
Both are memories for storing one frame or more.

Here, the general properties of the interframe difference signal will be explained.

FIG. 5 is a schematic diagram for explaining how inter-frame difference data is generated as an object moves between frames. The same figure (al is a figure showing the state one frame ago, and represents that the object S indicated by the diagonal line exists at 70 points. The figure (b) shows the current state. This is a diagram showing that the object S has moved from point 70 to point V3.

At this time, the inter-frame difference signal is expressed as the change obtained by subtracting (al) from the figure (b), as shown in figure (C). In other words, if the object moves between frames,
It can be seen that a large absolute value of the inter-frame difference signal occurs around the edge portion where the density change is large.

The contents of the difference memory unit 10 are the decoded values of the inter-frame difference signals of the previous frame, and for the reason just explained, it includes information regarding the edge position of the moving object. Therefore, if the movement of the target person (object) is relatively smooth (the movement does not change drastically between frames), such as in a TV conference image, the content of the difference memory section 10 will determine whether the object in the current frame is This means that the position of the edges of the body can be predicted.

Although it is possible to predict the value of the inter-frame difference signal itself depending on the contents of the difference memory section 10, even a shift of one pixel position around the edges may reverse the positive/negative of the difference value or cause the value to change. It has been experimentally confirmed that the value of the inter-frame difference signal may change significantly, and it is judged that it is difficult to predict the value itself of the inter-frame difference signal.

The mask creation unit 8 reads the contents of the portion corresponding to the current input block position from the difference memory unit 10 and creates a mask having the same size as the input block.

As described above, it is considered that the portion where the absolute value of the decoded value of the inter-frame difference signal is large corresponds to the edge portion, and a mask is created using the following procedure.

FIG. 6 is an explanatory diagram showing the procedure for creating a mask in the mask creating section 8. +al in the figure is the differential memory section 1
It shows the decoded value of the inter-frame difference signal in the previous frame read from 0, and the (i, j)th
The th pixel is expressed as d (i, j).

The figure (bl shows the shape and value of the mask created based on the decoded values in figure (a). The mask can be expressed as a two-dimensional array of the same size as the block. , the value of the (i, j)th pixel in the block is m(i, j)
I will express it as m(f.

As the value of j), the predicted position with respect to the edge portion is represented by 1, and the other values are represented by O. Here, whether or not it is an edge portion is determined, for example, as follows. That is, T
Assuming that h is a predetermined darkness value, ld (i,
j) If I≧Th, determine that d (i, j) is significant, and set m (i, j) = 1.

On the other hand, if ld (i, j)+<rh, then d (
i, j) is determined to be non-significant, and m (i, j)=O.

As a result, an array consisting of elements 1 and 0 is obtained as shown in FIG. It shows the predicted position for the position.

The same figure (C) shows that 1
This figure shows the result of enlarging one pixel area. The shape of the mask obtained in (b) of the same figure is generally linear, and if the mask position and the actual edge position deviate by even one pixel, there is a risk that the noise low frequency processing will not work effectively. Therefore, we performed an enlargement process on the 11-pixel region for the mask obtained in the same figure (bl) to give width to the predicted position of the edge part. For example, if the center pixel E in Figure 3 is "0" and there is even one "l" among the eight surrounding pixels, the "0" is changed to "1". It is sufficient to perform the loading process once (n≧1).

Finally, the mask processing section 9 will be explained. The mask processing unit 9 selectively performs noise reduction processing on the inverse orthogonal transform output based on the mask obtained by the mask creation unit 8. In other words, considering within a block, for the (i, j)th pixel f (i, j) of the inverse orthogonal transform output, when m (i, j) - 1 (area estimated to be an edge part) f ( i, j) →α+ Xf(t, J)m (i, j
) - 0 (region estimated to be flat) f (i, j)
→α. xf(i,j) For example, α, =1.0,
α. ==0.5 or process f (i, j) −→f (i, j).

As a result, it is possible to effectively reduce the above-mentioned noise component, which is particularly noticeable in the flat areas of shading, without blurring the edge areas.

The output that has been masked by the above procedure is sent to the branching unit 11 (branching unit 11 is omitted in Fig. 1, but since it is also used in the conventional configuration, its explanation will be omitted here) for differential processing. The data is written to the memory unit 10.

On the other hand, this output is added to the contents of the image memory section 14 to obtain a decoded output for the input image signal. In this way, the present invention reduces noise.

The above is the basic operation of the present invention. Next, some applied operations will be described.

First, in the above explanation, noise reduction processing is performed corresponding to the 1°0 value of each element in the mask, so if the contents of the mask are all "0", that is, there seems to be no edge part. Even for relatively flat portions, noise reduction processing is performed on the inverse orthogonal transform output. In a region where the mask contents are all 0'', there is little noise component generation (,
In order to improve the reproducibility of signals that have undergone orthogonal transformation, quantization, and inverse orthogonal transformation, we output the values as they are rather than applying the aforementioned noise reduction processing to the inverse orthogonal transformation output. It's better. Therefore, if there is no element for which m (i, j) = 1 in the created mask, α
. = 1.0, that is, the inverse orthogonal transform output is output as is.

Next, when a motion-compensated inter-frame difference signal is used as the inter-frame difference signal, motion compensation is also performed on the contents of the difference memory unit 10, thereby improving the accuracy of prediction for edge positions.

1 stored in the image memory for the human block
The motion vector between frames is obtained by searching for the most similar block from the decoded image of the previous frame and finding the relative positional shift. By reading the contents of the position shifted by the amount corresponding to this motion vector from the difference memory, it is possible to predict the edge position in a form commensurate with the movement of the object between frames, making the masking process easier. It becomes possible to carry out the process accurately.

The above explanation has been about the configuration on the transmitting side. Next, the configuration of the receiving side will be explained. FIG. 7 is a block diagram of a decoding method using the present invention on the receiving side. The quantized value from the transmitting side is decoded by the inverse orthogonal transform section 6, and the noise generated during decoding is reduced by the noise reduction processing section 7 as in FIG. The output of the noise reduction processing section 7 is added to the contents of the image memory section 14 to obtain the final received image. In addition,
The noise reduction method is the same as in the case of FIG. 4, and will be omitted here.

As described above, the present invention includes the noise reduction processing unit 7 to quantize the result of orthogonal transformation applied to the inter-frame difference signal or the motion-compensated inter-frame difference signal, and further obtain a decoded image by inverse orthogonal transformation. This effectively reduces noise components on images that occur when images are transmitted, making it possible to transmit moving image information at a low bit rate and improve the quality of received images.

(Specific Circuit Configuration) Next, the circuit configuration and operation of the noise reduction processing section 7, which is a feature of the present invention, will be described in detail. FIG. 8 is a block diagram of the noise reduction processing section 7, in which 8 is a mask creation section, 81 is an absolute value circuit, 82 is a determination circuit, and 83.degree. 84 is a value α1. α.

A terminal 85 is a switch for selecting 83.84 according to the output of the determination circuit 82. 9 is a mask processing section, 91 is a circuit for rearranging data from two-dimensional to one-dimensional data on the output of the inverse orthogonal transform section 6, and 92 is a multiplication circuit. 10 is a differential memory section, 101 is a differential memory main body, 102 is a block Bl, 1. 103 is a read control circuit for the memory 101, and 104 is a write control circuit for the differential memory 101.

Next, the operation of each part will be explained. To make the explanation easier to understand, reading of difference data from the difference memory 101 will be explained first. For example, consider the case where the (p, q)th block B Dr Q on the screen is processed.

FIG. 9 is a diagram showing a method for reading out the difference data stored in the difference memory 101 in units of blocks. The difference memory 101 stores difference data (inverse orthogonal transform unit 6
Assume that the result of performing mask processing on the output of ) is stored two-dimensionally. If the block size is MXN pixels, the first pixel address of B2.9 block is X++)
'+ is ((p -1)XM+L (q-1)XN+1
), and the last pixel address XMIYN is (pxM,
qxN). When reading the contents of block B9.9 from the difference memory 101, as shown by the arrow in the figure,
Difference data at each pixel position may be sequentially read out by horizontal scanning. At this time, in order to read the differential data within the block, the pixel address Xi+3'J (j=1,
---, M, j = 1. -・-N) is required. This calculation circuit is the intra-block pixel address calculation circuit 102 shown in FIG. Read control circuit 103
functions to read difference data from the difference memory 101 according to the pixel address Xi*yj sent from the intra-block pixel address calculation circuit 102.

Note that the difference memory 101. For the read control circuit 103 and the write control circuit 105, existing memory manufacturing techniques and memory control methods for two-dimensional image memories can be directly applied, so further explanation will be omitted.

FIG. 10 is a block diagram of the intra-block pixel address calculation circuit 102, where 110 and 111 indicate block positions p.

Terminals 112 and 113 are terminals for inputting q, terminals 121 and 12 are terminals for outputting pixel addresses x and y.
2 is a subtraction circuit, 123, 124 is a multiplication circuit, 125
, 126 is an adder circuit, 127 is a gate circuit, 128 is a pixel timing generation circuit, 129 and 130 are counters,
131 and 132 are memories, and 133 is a determination circuit. Block positions p, q obtained from input terminal 110°III
, the subtraction circuits 121 and 122 and the multiplication circuit 12
3 and 124, (p-1)xM, (q-1
)XN and write it to memory 131.132. Also, counters 129 and 130 are reset.

A pixel timing generation circuit 128 generates timing for each pixel, and a counter 129 is counted up thereby. The output of the counter 129 is added to the contents of the memory 131 in an adder circuit 125, and while the output of the counter 129 is less than or equal to M, it is sequentially sent to the output terminal 112 as a pixel address X via a gate circuit 127. On the other hand, it is determined whether the output of the counter 129 exceeds M or not at step i 133
If it exceeds M, the gate circuit 127 is closed, the counter 129 is reset, and the contents of the counter 130 are counted up. counter 13
The output of 0 is added to the contents of the memory 132 in an adder circuit 126 and sent to the output terminal 113 as a pixel address y. As described above, pixel addresses x and y necessary for sequentially reading out the contents of the (p, q)th block according to horizontal scanning are generated and sent to the readout control circuit 103.

Now, the contents read from the difference memory 101 in FIG. 8 (corresponding to the decoded value of the inter-frame difference signal in the previous frame) are sent to the mask creation section 8. In the mask creation section 8, first, the absolute value circuit 81 calculates the absolute value of the difference data. In the mask creation section 8, a constant α0. α. A determination circuit 82 determines whether the output of the absolute value circuit 81 is greater than or equal to the dark value Th, and based on the determination result, the switch 85 is connected to the terminals 83.84.
By connecting to one of the terminals 84, mask information (either α2 or α is selected for each pixel) is sent to the mask processing unit 9.

In the mask creation unit 8 in FIG. 8, the thickening process for the "1" pixel region within the mask is omitted, but an example of the configuration in which this is included is shown in FIG. 81 to 85 in the figure are the same as the contents of the mask creation section 8 in FIG. 801 is a pixel address (i, j) generation circuit, 802.803 is an addition circuit, 804.805 is a subtraction circuit, 806 is an OR circuit, 807 to 811 is a read control circuit for the memory 821, and 820 is a write control circuit for the memory 821. circuit,
821 is a memory for storing mask information obtained from the switch 85 (before thickening, α1=1.α.=0, and the mask is expressed by data of 1.0); 822 is a write control circuit for the memory 823, 823 is a memory for storing mask information after thickening processing, 824 is a read control circuit from the memory 823, 825 is a determination circuit, 826 and 827 are constants α'1.
α′. A terminal 828 for reading out is a switch for selecting terminals 826 and 827 according to the output of the determination circuit 825.

Next, the operation of each part will be explained. The mask information obtained by the mask creation section 8 is temporarily written into the memory 821 via the write control circuit 820. The pixel address generation circuit 801 is configured in accordance with FIG.
Generate a pixel address (1+j) for 1. For pixel address (i, j), adder circuit 802.803 and subtracter circuit 804,805 are used to calculate (i+1.j),
(i-1, j), (i, j +1),
Generate a set of addresses (i, j-1). (i
, j), the readout control circuit 8
The mask information is read from the memory 821 through steps 07 to 811, and an OR output of these is obtained in an OR circuit 806.

As a result, when (i, j) is "1", the output is "l", and when (+, j) is "0", one of the four neighbors is "1".
If so, an output "1" is obtained, and if all are "0", an output "0" is obtained, and in the end, fattening processing for one pixel can be realized. The result of the thickening process is written to the memory 823 via the write control circuit 822. The contents of the memory 823 are sequentially read out via the read control circuit 824, and
5, 1.0 is determined. Based on this result, the switch 828 is controlled, and the constant α'1 or α'' is successively output from the terminal 826 or 827 and sent to the mask processing section 9 (specifically, the multiplication circuit 92).

Next, the mask processing section 9 shown in FIG. 8 will be explained.

Reference numeral 91 denotes a data rearrangement circuit for rearranging data from two-dimensional to one-dimensional data on the two-dimensional inverse orthogonal transform output obtained by the inverse orthogonal transform unit 6.

For this output, using the multiplier circuit 92, the switch 85
Constant α, or α, sent via . (In the case of FIG. 11, the constant α” sent through the switch 828,
Or α゛. ). This gives us a constant α.

(In the case of FIG. 11, α゛.) For the pixels, the value of the inverse orthogonal transform output is forcibly reduced, and as described above, 1. As a result, it is possible to reduce noise components in the flat gray area. Although the block diagram is omitted, when the output of the determination circuit 82 is 7830, the value of the output pixel of the inverse orthogonal transform section 6 is sent as is to the branching section 11, while when the output is No, for example, the average value of nine neighboring pixels is calculated. Smoothing processing may also be performed by sending the data to the branching unit 11.

The output of the mask processing section 9 passes through the branching section 11 and then the adding section 13.
At the same time, the data is sent to the difference memory section 10 and written to the difference memory 101 via the write control circuit 104. The intra-block pixel address at this time is given by the intra-block pixel address calculation circuit 102.

(Effects of the Invention) As described above, by providing the noise reduction processing unit 7 of the present invention, moving image information can be transmitted at a low bit rate using orthogonal transform encoding for an inter-frame difference signal or a motion-compensated inter-frame difference signal. It is possible to effectively reduce the noise component on the differentially decoded signal, which is a problem when decoding. In particular, a mask is created to distinguish between edge parts and light and shade flat parts based on the decoding results of inter-frame differences in the previous frame.
By selectively performing smoothing processing, it is possible to reduce noise components that are noticeable in flat areas without blurring edge areas.

In particular, the noise component discussed here is a major factor in deteriorating image quality, and is an obstacle to putting the above-mentioned encoding system into practical use. According to the present invention, since it is possible to reduce noise components, it is possible to realize moving image transmission at a low bit rate using the above-mentioned encoding method and to improve the quality of received images, and the effects thereof are extremely large.

[Brief explanation of drawings]

Figure 1 is a configuration diagram of a conventional orthogonal transform encoding method for inter-frame difference signals or motion-compensated inter-frame difference signals;
The figure is a block diagram of an orthogonal transform coding system incorporating conventional noise reduction processing, Figure 3 is a diagram showing a 3x3 pixel block for explaining the conventional noise reduction method, and Figure 4 is a diagram according to the present invention. A block diagram of the orthogonality transform encoding method, FIG. 5 is a schematic diagram for explaining how inter-frame difference data is generated, and FIG. 6 is an explanation for explaining the procedure of the mask creation unit 8 used in the present invention. 7 is a block diagram showing an embodiment in which the present invention is applied to the receiving side, FIG. 8 is a block diagram of the noise reduction processing unit 7 used in the present invention, and FIG. 9 is a block diagram showing the configuration of the noise reduction processing unit 7 used in the present invention. Diagram 1 for explaining the operation of accessing the differential data stored in the block in block units.
FIG. 0 is a block diagram of the in-block pixel address calculation circuit 102 used in the present invention, and FIG. 11 is a block diagram of the mask creation unit 8 including the enlargement process of the 11-pixel area used in the present invention. Voice 50 S' (01(b) (c) Yuu 74

Claims (1)

[Claims] Difference obtained by performing orthogonal transformation on the difference signal between frames or motion compensation frames of a moving image signal obtained by scanning a moving image to be transmitted, quantizing the result, and then inverse orthogonal transforming the quantized output. an input terminal that receives the decoded signal, and outputs a differential decoded signal in which noise contained in the differential decoded signal is reduced by selectively performing smoothing processing on the differential decoded signal based on a mask that is updated every frame. a mask processing means for storing the noise-reduced differential decoded signal, a differential memory means for accumulating at least one frame worth of differential decoded signals, and increasing the differential signal by using the differential decoded signal stored in the differential memory means. a mask creating means for creating the mask for each frame and supplying the mask to the mask processing means so as to make the predicted position on the moving image to be transmitted significant, the output side of the mask processing means; A noise reduction method for a differentially decoded signal in moving image transmission, which is configured to extract a differentially decoded signal with reduced noise.