JPH08294128A - Image coding decoding method - Google Patents

Abstract

PURPOSE: To eliminate a noise component included in a decoded image efficiently. CONSTITUTION: An obtained decoded image 111 is given to a time space low pass filter 121, in which the image is divided into the same small blocks as those at a transmitter side, and either of data of a small block of a current frame or a difference between the data of the small block of the current frame and data 132 by a motion compensation section 131 from a small block designated by a motion vector 106 from decoded image data of a preceding or a succeeding frame stored in a frame memory 129 are selected by an inter-frame/in-frame selection flag 105. An orthogonal conversion section 124 applies orthogonal transformation to the selected data and whether or not the obtained orthogonal transformation coefficient is in existence in a range 117 decided by a range setting section 116 according to a quantization level 115 is discriminated, and when the orthogonal transformation coefficient is not present in the range 117, a clipping section 125 is used to clip the coefficient so as to be resident in the range.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a high-efficiency coding / decoding method using motion-compensated interframe prediction orthogonal transform.

[0002]

2. Description of the Related Art As a highly efficient image coding method, there is a method using an orthogonal transform represented by a discrete cosine transform.

FIG. 3 is a block diagram showing a general orthogonal transform coding / decoding method. First, the image 202 input from the input terminal 201 is N ×
It is divided into N small blocks. The orthogonal transformation unit 204 performs orthogonal transformation on each of the divided small blocks, and the obtained orthogonal transformation coefficient is quantized with the quantization characteristic determined by the quantization unit 205. Next, the code allocation unit 20
A code is assigned to the quantized representative value at 6, and the encoded data 207 is transmitted to the receiving side through the output terminal 208. On the other hand, on the receiving side, the code decoding unit 211 decodes the code from the coded data 210 received at the receiving terminal 209, and the inverse quantization unit 212 reproduces the quantized representative value for each orthogonal transform coefficient, and the inverse Orthogonal transformation unit 213
Inverse orthogonal transformation is performed in and output to the output terminal 215 as the decoded image data 214.

The quantization method is generally as shown in FIG.
When the input value F is F _i ≦ F <F _{i + 1} , the quantized representative value is F
_i ′ (where F _i ≦ F _i ′ <F _{i + 1} ) is set.

In such an orthogonal transform coding / decoding method, since coding is performed in units of small blocks, when the compression rate is set high, the quantization is rough and block noise is generated in the decoded image. There is a problem that deterioration of image quality such as the above is detected. Various methods have been considered to avoid this, but multiple constraint conditions that the original image must satisfy must be set, and the decoded image that has deteriorated due to encoding must be repeatedly corrected to satisfy all these constraint conditions. Then, as a result, there is a method to restore the original image as close as possible (A. Zakhor, “Iterative procedures f
or reduction of blocking effects in transform image
coding, ”IEEE Trans. on Circuits and Systems for V
ideo Technology, vol. 2, No.1, pp.91-95, Mrch 199
2).

In the method in this document, the constraint condition is
The condition that the original image is smooth and the condition that the range in which the true value before quantization is present are limited from the quantized representative value are used. The former is achieved by applying a low pass filter to the image. In the latter, when the quantized representative value is F _i ′, the true value F before quantization is F _i ≦ F
Since it must exist in <F _{i + 1} , it may be modified to satisfy this.

Specifically, first, the decoded signal obtained on the receiving side is
A low-pass filter is applied to the signal. Next, low pass
The filtered signal is orthogonally transformed. Obtained at this time
Orthogonal transformation coefficient F_L Is the signal before applying the low pass filter
Coefficient F obtained by orthogonal transformation _i'Not always after
Does not meet the requirements of the person. Therefore, the following conditions
By F_L To fix.

[0008]

[Equation 1] That is, when the value exceeds the true value range, the clipping process is performed to the closer one of the maximum value and the minimum value of the true value range. The orthogonal transform coefficient corrected by clipping is subjected to inverse orthogonal transform to obtain corrected image data. A final corrected image is obtained by repeating the above low-pass filter, orthogonal transformation, clipping, and inverse orthogonal transformation a predetermined number of times.

FIG. 5 is a block diagram for executing the above-described method for correcting a decoded image. On the receiving side, the code decoding unit 30 uses the encoded data 302 received at the receiving terminal 301.
3, the inverse quantization unit 304 reproduces the quantized representative value for each orthogonal transform coefficient, and the inverse orthogonal transform unit 305 performs the inverse orthogonal transform to obtain the decoded image data 30.
6 is obtained. The decoded image data 306 is filtered by the low-pass filter unit 307, then orthogonally transformed by the orthogonal transformation unit 308, and the obtained orthogonal transformation coefficient is input to the clipping unit 309. In the clipping unit 309, the orthogonal transform coefficient is modified by using the maximum and minimum values that can be obtained by the range setting unit 310 before quantization by the quantized representative value used in the inverse quantization unit 304. The corrected orthogonal transformation coefficient is subjected to inverse transformation in the inverse orthogonal transformation unit 311, and the first corrected image 312 is obtained. The corrected image 312 is subjected to the second low-pass filter in the low-pass filter section 313, the orthogonal transformation is performed in the orthogonal transformation section 314, and the obtained orthogonal transformation coefficient is input to the clipping section 315. In the clipping unit 315, the orthogonal transform coefficient is modified similarly to the first time, and the modified orthogonal transform coefficient is subjected to the inverse transform in the inverse orthogonal transform unit 316 to obtain the modified image 317 for the second time. Similarly, the n-th corrected image 318 obtained by repeating the process a predetermined number of times
Is finally output to the output terminal 319.

[0010]

Consider a case in which the above-described method is applied to an image coded / decoded by motion compensation interframe differential orthogonal transform in which coding is also performed using interframe correlation.

FIG. 6 shows a block configuration on the transmission side for performing orthogonal transform coding on the motion compensation inter-frame difference signal. First, the image data 402 of the current frame input from the input terminal 401 is divided into small blocks, and the divided small blocks are moved from the image data of the past or future frames stored in the frame memory 403. The most similar block is calculated according to the motion vector 405 calculated by the vector detection unit 404. According to the obtained motion vector 405, the motion compensation unit 406 obtains the motion compensation prediction data 407 from the image data of the past or future frame. A difference unit 408 calculates a difference value 409 between motion compensation frames using the small block data 402 of the current frame and the motion compensation prediction data 407. After that, either the inter-frame difference value 409 of the small block or the small block pixel value 402 itself in the current frame is selected, and the orthogonal transformation unit 410 performs orthogonal transformation on the selected data for each small block. The obtained orthogonal transform coefficient is quantized by the quantization step width determined by the quantization unit 411 to obtain a quantization level number, and the code assignment unit 412 assigns a code corresponding to the quantization level number and transmits it. A code is also assigned to the motion vector 405 and the inter-frame / intra-frame selection flag 417, which is transmitted as data 413 together with the orthogonal transform coefficient coded data.

On the other hand, the quantization level number is calculated by the inverse quantization unit 41.
Inverse quantization is performed at 4 to obtain a quantized representative value, and then inverse orthogonal transformation is performed at the inverse orthogonal transformation unit 415. The data which has been subjected to the inverse orthogonal transform has the inter-frame / in-frame selection flag 4
If 17 is within the frame, the frame memory 4
If the selection flag 417 is between frames, the motion compensation prediction data 407 and the adder 416 are written.
Are added and then written in the frame memory 403. These processes are also performed on the receiving side.

As described above, in the motion-compensated inter-frame difference orthogonal transform coding, the motion-compensated inter-frame difference value is orthogonally transformed and quantized / encoded in a region where motion compensation is appropriately performed. In a region where the inter-frame difference value is large even if motion compensation is performed, the pixel values in the frame are orthogonally transformed and quantized / encoded.
For the latter, the conventional method can be applied as it is because it can be regarded as intra-frame encoding, but in the region where the motion-compensated inter-frame difference signal is quantized like the former, what can be known on the decoding side is , The true value range of the orthogonal transform coefficient of the motion compensation inter-frame difference signal, and not the true value range of the orthogonal transform coefficient of the pixel value itself in the frame. Therefore, it is not appropriate to directly clip the coefficients obtained by orthogonal transformation of the image signal after the low-pass filter, and it is possible to obtain an appropriate corrected image only by repeating the low-pass filter, orthogonal transformation, clipping, and inverse orthogonal transformation. I can't. For this reason, there is a problem that it is difficult to prevent deterioration on the decoded image.

An object of the present invention is to provide an image decoding method capable of efficiently removing noise components contained in a decoded image.

[0015]

According to the image coding / decoding method of the present invention, a digital image signal such as a television signal is divided into small blocks on the transmitting side by dividing the image data of the current frame. For the small blocks, the most similar block is selected from the image data of the past or future frames stored in the frame memory, and the motion compensation interframe difference value with the small block of the current frame is calculated. Select either the inter-frame difference value of the block or the small block pixel value itself in the current frame, and determine the orthogonal transform coefficient obtained by performing the orthogonal transform for each small block on the selected data. To obtain a quantized representative value by quantizing with a quantization step width, assign a code corresponding to the quantized representative value, and calculate an interframe difference. It is transmitted together with a motion vector for specifying a similar block and an inter-frame / intra-frame selection flag. On the receiving side, the transmitted code is decoded to reproduce the quantized representative value, and the reproduced quantized representative After performing the inverse orthogonal transform on the value, the inter-frame / intra-frame selection flag causes the value itself after the inverse orthogonal transform, or the value of the small block specified by the motion vector from the past or future decoded frames. In an image coding / decoding method for obtaining a decoded image by using one of the values after the inverse orthogonal transformation as the decoded value of the small block, a space-time low-pass for the decoded image obtained at the receiving side. The process of applying a filter, dividing the image signal after the spatiotemporal low-pass filter into the same small blocks as those used on the transmission side, and Data of a small block of the frame, or a divided small block of the current frame, and the difference value between the small block specified by the motion vector from the decoded image data of the past or future frame stored in the frame memory. A process of selecting either of the data by the inter-frame / intra-frame selection flag, a process of performing an orthogonal transform on the selected data, and whether or not the obtained orthogonal transform coefficient exists within a predetermined range. Determination processing, if not present, clipping processing within the range, processing for performing inverse orthogonal transformation on the orthogonal transformation coefficient subjected to clipping processing, and inverse orthogonal transformation by the inter-frame / intra-frame selection flag. The later value of that, or of the past or future decoded frames specified by the motion vector. There is a process to obtain a modified image by making the modified value of the small block one of the value obtained by adding the value after the inverse orthogonal transformation to the small block value. It is characterized in that the final corrected image is obtained by repeatedly performing the processing of (3) a predetermined number of times.

[0016]

According to the present invention, when the pixel itself in a frame is subjected to orthogonal transform coding, clipping is performed on the intraframe orthogonal transform coefficient, and when the motion compensation interframe difference value is subjected to orthogonal transform coding, the motion compensation is performed. Clipping is applied to the orthogonal transform coefficient of the inter-compensation frame difference signal.

[0017]

Embodiments of the present invention will now be described with reference to the drawings.

FIG. 1 is a block diagram of an image coding / decoding apparatus according to an embodiment of the present invention. With respect to the code string 102 received at the receiving terminal 101, first, the code decoding unit 103 decodes the transmitted code to quantize level number 1
04, the inter-frame / intra-frame selection flag 105, and the motion vector 106 are reproduced. The reproduced quantization level number 104 is inversely quantized by the inverse quantization unit 107 to obtain a quantized representative value, and then inversely orthogonally transformed by the inverse orthogonal transformation unit 108, and then the interframe / intraframe selection flag 105 is used. The switch 110 is switched, and the motion compensating unit 1 is selected from the value itself after the inverse orthogonal transform or the past or future decoded frame stored in the frame memory 112.
At 13, the motion-compensated prediction data 114 designated by the motion vector 106 is added with the value after the inverse orthogonal transform by the adder 109, and one of them is the decoded value 11 of the small block.
Set to 1.

The obtained decoded image 111 is input to the clipping processing block 120 ₁ and subjected to the spatiotemporal low-pass filter 121. The image signal after the spatiotemporal low-pass filter 121 is divided into the same small blocks as those used on the transmitting side, and stored in the small block data of the current frame, or the divided small block of the current frame and the frame memory 129. The difference between the small block specified by the motion vector 106 and the data 132 extracted by the motion compensation unit 131 from the decoded image data of the past or future frame, which is obtained by the difference unit 122, It is selected by the switch 123 according to the in-frame selection flag 105. The selected data is subjected to orthogonal transformation by the orthogonal transformation unit 124, and the obtained orthogonal transformation coefficient is set to the clipping range 117 determined by the range setting unit 116 by the quantization level 115 obtained by the inverse quantization unit 107. Exists in the. In the clipping unit 125, if the orthogonal transform coefficient does not exist in the set range, clipping processing is performed within the set range.

The orthogonal transform coefficient subjected to clipping processing is subjected to inverse orthogonal transform in the inverse orthogonal transform unit 126, and the inter-frame / intra-frame selection flag 105 is used to determine the value itself after the inverse orthogonal transform, or the past or future decoding. The value after the inverse orthogonal transform is added to the value 132 of the small block designated by the motion vector 106 in the frame by the adder 127.
Either one of the values added in step 1 is selected by the switch 128 and the corrected value of the small block is selected to obtain the corrected image 130. For the corrected image 130 obtained, the final image 13 from the spatiotemporal low-pass filter 121 described above
The final corrected image 133 is obtained at the output terminal 140 of the clipping processing block 120 _n by repeatedly performing the processing until 0 is obtained a predetermined number of times.

Specific examples of the present invention are shown below. In this specific example, consider the case where discrete cosine transform is used for a motion-compensated inter-frame difference signal as an information compression method.

On the transmitting side, the input image is divided into small blocks of 8 × 8 f (i, j) (i, j = 0,1,2, ...
7) and the motion-compensated inter-frame difference signal e (i, j) (i, j, = 0,1,2, ...
・, 7) is obtained. For the original signal f (i, j) in the frame and the inter-frame difference signal e (i, j), the values defined by the following equations are calculated.

[0023]

[Equation 2] When VAROR> VAR, the intra-frame / inter-frame selection flag P (105) is set to 1, and VAROR is reversed.
If ≦ VAR, the in-frame / inter-frame selection flag P is set to 0. When P = 0, f (i, j) is subjected to discrete cosine transform to obtain a transform coefficient C (i, j). When P = 1, the discrete cosine transform is performed on e (i, j) to obtain the transform coefficient C (i, j).

The obtained conversion coefficient is as shown in FIG.
Quantization is performed by a quantizer having a step width of 2d. At this time, if the value C (i, j) before quantization is 2nd−d ≦ C (i, j) <2nd + d, the quantized representative value C ′ (i.j) is C ′ (i, j). j) = 2nd. In this case, on the contrary, C '(i, j) = on the receiving side.
When 2nd is transmitted, the true value C (i, j) of the DCT coefficient before quantization is 2nd−d ≦ C (i, j) <2.
It can be seen that it is in the range of nd + d.

Now, consider a case where the quantized representative value C '(i, j) obtained on the receiving side is 2nd.

First, when the intra-frame / inter-frame selection flag P is P = 0, C '(i, j) is subjected to inverse discrete cosine transform to obtain a decoded image f ⁰ (i, j). When the intra-frame / inter-frame selection flag P is P = 1, after the inverse discrete cosine transform of C ′ (i, j), the area specified by the motion vector 106 in the decoded image of the previous frame. And the decoded image f ⁰
(I, j).

First, the following spatiotemporal low-pass filter 121 is applied to the obtained decoded image f ⁰ (i, j).
Give.

[0028]

(Equation 3) However, g ⁰ (i, j) and h ⁰ (i, j) are f ⁰ (i, j)
j) represents a decoded image one frame before and one frame after. Also, the low-pass filter is applied across the boundaries of the small blocks.

The image signal subjected to the low-pass filter is converted into 8
When divided into × 8 small blocks and the intra-frame / inter-frame selection flag P is P = 0,

[0030]

[Outside 1] The orthogonal transform unit 124 performs a discrete cosine transform on
Conversion factor

[0031]

[Outside 2] Get. When P = 1,

[0032]

[Outside 3] And the area in the previous frame specified by the motion vector 106

[0033]

[Outside 4] The orthogonal transform unit 124 performs a discrete cosine transform on
Conversion factor

[0034]

[Outside 5] Get.

Obtained conversion coefficient

[0036]

[Outside 6] On the other hand, the clipping unit 125 performs the following clipping processing.

[0037]

[Equation 4] That is, when the value exceeds the range of the true value, it means that the true value is clipped to the maximum value or the minimum value, whichever is closer.

After the transform coefficient subjected to the clipping process is subjected to the inverse discrete cosine transform in the inverse orthogonal / transform unit 126, if the intra-frame / inter-frame selection flag P is P = 0,
The corrected image f ¹ (i, j) for the first time is used as it is. When the intra-frame / inter-frame selection flag P is P = 1, the transform coefficient subjected to the clipping process is subjected to the inverse discrete cosine transform and then added to the data 132 of the region of the previous frame indicated by the motion vector 106. The added value is added by the device 127 to obtain the first corrected image f ¹ (i, j).

The above processing is repeated n times to obtain the n-th corrected image f ⁿ (i, j) as the final output 133.

[0040]

As described above, according to the present invention,
When the pixel itself in the frame is orthogonal transform coded, clipping is performed on the intraframe orthogonal transform coefficient, and when the motion compensation interframe difference value is orthogonal transform coded, the motion compensation interframe difference value signal By clipping the orthogonal transform coefficients, the corrected image can be approximated to the original image before encoding, and the noise component included in the decoded image can be efficiently removed.

[Brief description of drawings]

FIG. 1 is a block diagram of an image encoding / decoding device according to an embodiment of the present invention.

FIG. 2 is a diagram showing a quantization characteristic in the embodiment of FIG.

FIG. 3 is a block diagram of a general orthogonal transform encoding / decoding device.

FIG. 4 is an explanatory diagram showing an orthogonal transform coefficient quantization method used for orthogonal transform coding.

FIG. 5 is a block diagram of a conventional decoding device that corrects a decoded image by repeating a low-pass filter and orthogonal transform coefficient clipping.

FIG. 6 is a block diagram of a motion compensation inter-frame difference value orthogonal transform coding apparatus.

[Explanation of symbols]

101 Reception Terminal 102 Received Coded Data 103 Code Decoding Unit 104 Quantization Level Number 105 Interframe / Intraframe Selection Flag 106 Motion Vector 107 Inverse Quantization Unit 108 Inverse Orthogonal Transform Unit 109 Adder 110 Switch 111 Decoded Image 112 Frames Memory 113 Motion Compensation Unit 114 Motion Compensation Prediction Data 115 Quantization Level 116 Range Setting Unit 117 Set Clipping Range 120 _{1 to} 120 _n Clipping Processing Block 121 Spatio-Temporal Low-Pass Filter 122 Difference Unit 123 Switch 124 Orthogonal Transformation Unit 125 Clipping Unit 126 Inverse orthogonal transformation unit 127 Adder 128 Switch 129 Frame memory 130 First corrected image 131 Motion compensation unit 132 Motion compensation prediction data 133 Final corrected image 140 Output Terminal

Claims (1)

[Claims] 1. For a digital image signal such as a television signal, the transmitting side divides the image data of the current frame into small blocks, and the divided small blocks are stored in a frame memory in the past. Alternatively, after selecting the most similar block from the image data of the future frame and calculating the motion compensation frame difference value with the small block of the current frame,
Select either the inter-frame difference value of the small block or the small block pixel value itself in the current frame,
Orthogonal transform coefficients obtained by performing orthogonal transform for each small block on the selected data are quantized with a predetermined quantization step width to obtain a quantized representative value, which corresponds to the quantized representative value. A code is assigned and transmitted together with a motion vector for specifying a similar block for calculating an inter-frame difference and an inter-frame / intra-frame selection flag. The reproduced representative value is reproduced, and the reproduced quantized representative value is inverse-orthogonal-transformed, and then the value itself after inverse-orthogonal transformation or the past or future decoded frame is selected depending on the inter-frame / intra-frame selection flag. From the value obtained by adding the value after the inverse orthogonal transformation to the value of the small block specified by the above motion vector, the decoded value of that small block is used for decoding. In the image coding / decoding method for obtaining an image, the same process as that used for the transmission side is the process of applying the spatiotemporal low-pass filter to the decoded image obtained on the receiving side It is specified by the motion vector from the process of dividing into small blocks, the data of the small blocks of the current frame, or the divided small blocks of the current frame, and the decoded image data of past or future frames stored in the frame memory. Processing of selecting either of the data of the difference value with the small block by the inter-frame / intra-frame selection flag, the processing of performing the orthogonal transformation on the selected data, and the obtained orthogonal transformation coefficient. The process of determining whether it exists in the specified range, the process of clipping in the range if it does not exist, and the clear A process of performing an inverse orthogonal transform on the orthogonal transform coefficient subjected to the wrapping process, and
Depending on the intra-frame selection flag, either the value after the inverse orthogonal transform itself or the value obtained by adding the value after the inverse orthogonal transform to the small block value specified by the motion vector from the past or future decoded frames. The process of obtaining a corrected image is performed by using or as the correction value of the small block, and the final corrected image is obtained by repeatedly performing all the above-mentioned processes on the obtained corrected image for a predetermined number of times. An image encoding / decoding method characterized by obtaining.