JP3186685B2 - Motion compensation prediction method - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a motion compensation prediction method used for a video conference, a videophone, a cable television, and the like.

[0002]

2. Description of the Related Art FIG. 5 is a block diagram showing a configuration example of a conventional motion-compensated inter-frame predictive coding apparatus. In FIG. 5, a digitized signal (image) is input to an adder 502 from an input terminal 501. You.

An adder 502 is provided with a frame memory 510.
The output value is also input and outputs a prediction error signal. Here, the prediction value is a prediction value obtained by motion compensation prediction.
You. Therefore, the output from frame memory 510 is
The measured value follows the motion vector detected by the motion vector detector.
Is the motion compensation predicted value (IEICE 19
1981 Vol.J64-B, No.1 P.24-31
Motion correction prediction method)). Specifically, the motion vector
Is 0, the predicted pixel of the input frame and the
Pixel value on the frame memory at the same position
Shall be prime. The same applies when the motion vector is other than 0.
The position of the predicted pixel in the input frame and the position in the frame
Pixel position corresponding to the value of the motion vector with the same position as 0
Pixel is a predicted pixel. In this case, the frame memory 5
A pixel is located at a pixel position corresponding to the value of the motion vector within 10.
If there is no pixel, interpolate from the pixels around the corresponding pixel position
The pixel at the corresponding pixel position is calculated by
(1990 Image Coding Symposium (PCSJ90) P.175-
177 (Detection of adaptive line interpolation inter-field motion compensation method
Discussion)). Thus, the motion vector is
It is used to indicate the relative position of the element. Soshi
Then, the prediction error signal is input to the orthogonal transform unit 503 and is orthogonally transformed. The transform coefficient is input to the quantization unit 504 and is quantized. The quantized transform coefficients are input to the inverse quantization unit 507 and the multiplexing unit 505. Multiplexer 5
Reference numeral 05 multiplexes the transform coefficient quantized by the quantization unit 504 and the motion vector detected by the motion vector detection unit 511, and outputs the multiplexed signal through an output terminal 506.

On the other hand, an inverse quantization section 507 inversely quantizes the quantized transform coefficient to obtain an orthogonal transform coefficient, and outputs this to an inverse orthogonal transform section 508. The inverse orthogonal transform unit 508 performs inverse orthogonal transform and outputs an original prediction error signal. The prediction error signal obtained by the inverse orthogonal transform is provided to the adder 509. The adder 509 adds the prediction value from the frame memory 510 to the prediction error signal from the inverse orthogonal transform unit 508, and outputs a reproduced pixel value to the frame memory 510. The frame memory 510 creates a reproduced image from the reproduced pixel values from the adder 509, and outputs the reproduced image as a predicted value of the next frame. Further, the motion vector detection unit 511 obtains a motion vector from the input image input from the input terminal 501 and the reproduced image from the frame memory 510, and outputs the motion vector to the frame memory 510 and the multiplexing unit 505.

[0005]

However, in the conventional motion-compensated inter-frame predictive coding apparatus, since a predicted image is formed from a frame image of a previous frame, a motion of an odd number of pixels in the vertical direction in one frame time is required. At
It is difficult to accurately detect a motion vector, and even if it is performed, an image of a different field is used as a predicted image, so that a highly accurate predicted image cannot be obtained, and the compression effect is deteriorated. .

SUMMARY OF THE INVENTION The present invention is directed to solving the above-mentioned problems of the prior art, and is capable of preventing deterioration due to a time difference for different inter-field predictions and providing an excellent motion compensation capable of obtaining an appropriate predicted image. The purpose is to provide a prediction method.

[0007]

In order to achieve this object, a motion compensated prediction method according to the present invention comprises a first motion vector between a reference frame and an encoded frame and a first field of the reference frame. And a second motion vector generated by multiplying the first motion vector by a predetermined ratio and a second field of the reference frame to perform motion compensation on the first field of the encoded frame. In the prediction method, the predetermined ratio is determined by a ratio between a time interval of a first motion vector and a time interval of a second motion vector.

Further, a second motion vector generated by using a first motion vector between the reference frame and the frame to be encoded and a second field of the reference frame and multiplying the first motion vector by a predetermined ratio is used. Using the motion vector and the first field of the reference frame, the second
A motion compensation prediction method for performing motion compensation on a field, wherein the predetermined ratio is determined by a ratio between a time interval of a first motion vector and a time interval of a second motion vector. Things.

According to the above method, the field of the predicted image always coincides with the field of the input image, and a predicted image from the same field is created even in the case of a motion of odd pixels in one frame in the vertical direction. This makes it possible to smoothly detect a motion vector without causing any positional deviation. Further, a more appropriate predicted image can be obtained, and the compression efficiency can be improved.

[0010] Furthermore, not only the field that matches the field of the predicted image but also the other field image is corrected by the motion vector and used as an image of the same field, so that the resolution of the predicted image can be increased. It is possible to output a predicted image with high accuracy, thereby improving the compression effect and improving the image quality.

[0011]

Embodiments of the present invention will be described below in detail with reference to the drawings.

(Embodiment 1) FIG. 1 is a block diagram showing the configuration of Embodiment 1 of the present invention, and FIG. 2 shows the principle of field image division and oversampling.

An input image (interlaced television) signal input from an input terminal 101 is input to a motion vector detector 113 and an adder 102, respectively. The adder 102 outputs the difference from the predicted image output from the frame combining unit 114 to the orthogonal transform unit 103,
Performs, for example, an orthogonal transform such as DCT, and outputs transform coefficients to a quantization unit 104. The quantization unit 104 quantizes the transform coefficients and outputs them to an inverse quantization unit 107 and a multiplexing unit 105, respectively. The multiplexing unit 105 outputs the motion vector detection unit 11
Then, the motion vector detected in Step 3 and the quantized transform coefficient are multiplexed, encoded, and output through the output terminal 106.

The inverse quantization unit 107 inversely quantizes the transform coefficient and outputs the result to the inverse orthogonal transform unit 108.
8 is converted into the original prediction error signal and output to the adder 109. The adder 109 adds the predicted image output from the frame synthesizing unit 114 to the field image dividing unit 1.
The reproduced image is output to 10.

The field image dividing section 110 divides the reproduced image into two fields, a first field and a second field, and outputs the divided fields to the oversampling section 111 and the oversampling section 112, respectively. Then, the oversampling unit 111 and the oversampling unit 112 respectively perform the interpolation of the input field image, output to the image memory 115 and the image memory 116, respectively, and store the same in the memory.

Image memory 115 and image memory 116
Is used to convert each field image of the predicted image based on the motion vector detected by the motion vector detection unit 113 to the frame synthesis unit 1
14, and is synthesized into a frame and output to the adders 102 and 109 as predicted images.

Further, the image memory 115 and the image memory 116 output the reproduced image interpolated to detect the motion vector to the motion vector detection unit 113, and the motion vector detection unit 113 transmits the reproduced image from the two image memories 115 and 116. A motion vector is detected from the interpolated reproduced image and the input image from the input terminal 101. Here, at the time of pattern matching performed for detecting a motion vector, the motion vector detection unit 113 performs matching of a corresponding field of the input image with a reproduction field image. Also, the image memories 115 and 1
Reference numeral 16 outputs only a predicted pixel of a corresponding field of the predicted image as a predicted image.

FIG. 2 shows the operation of field image division and oversampling. Field image division unit 1
The frame signal (a) input to 10 is divided into a first field and a second field alternately line by line (b). Since the number of lines of the divided field image is の of the original frame, if the accuracy of the motion vector is to be obtained up to one pixel unit, the number of lines must be the same as that of the original frame. Must.

Therefore, the oversampling unit 11
At 1,112, the pixel at the pixel position in the other field is obtained from the surrounding pixels by interpolation (c). The detection of the motion vector is performed by matching with a field image corresponding to the field of the input image. As a result, the fields of the input image and the fields of the predicted image can always be correlated with each other, so that the fields are not reversed, and the predicted image with a smaller prediction error can be output.

(Embodiment 2) FIG. 3 is a block diagram showing the configuration of Embodiment 2 of the present invention. An input image (interlaced television signal) input from an input terminal 301 is input to an adder 302 and a motion vector. It is input to the detection unit 313. The adder 302 outputs the difference between the predicted image output from the pixel interpolation unit 312 and the input image to the orthogonal transformation unit 303. This difference is called a prediction error signal.

This prediction error signal is converted by the orthogonal
An orthogonal transform such as CT is performed, and the transform coefficient is quantized by a quantization unit 304.
, And the quantization unit 304 quantizes the input transform coefficient, and quantizes the transform coefficient by the multiplexing unit 305 and the inverse quantization unit 307.
Respectively. Then, the multiplexing unit 305 multiplexes the motion vector detected by the motion vector detection unit 313 and the quantized transform coefficient, and outputs the multiplexed signal through the output terminal 306.

The inverse quantization unit 307 inversely quantizes the quantized transform coefficient to obtain an orthogonal transform coefficient, and outputs the orthogonal transform coefficient to the inverse orthogonal transform unit 308. The inverse orthogonal transform unit 308 performs an inverse orthogonal transform on the input orthogonal transform coefficient to obtain a prediction error signal. This prediction error signal is input to the adder 309 and added to the prediction image output from the pixel interpolation unit 312 to obtain a reproduced image. This reproduced image is stored in the image memory 310.

The image memory 310 detects the first motion vector, and outputs the reproduced image stored in the image memory 310 to the field adjustment unit 311 for predicting the next frame. The field adjustment unit 311 performs position compensation based on the time difference of the field with the second motion vector based on the motion vector detected by the motion vector detection unit 313, and outputs the result to the pixel interpolation unit 312. This pixel interpolation unit 31
2 calculates the pixels of the actual predicted image based on the image output from the field adjustment unit 311 and adds the predicted image to the adder 302.
And to the adder 309.

FIG. 4 shows the operation of the field adjustment unit 311 in FIG. When the reproduced image stored in the image memory 310 (hereinafter referred to as a reference frame image) is composed of a first field and a second field, and the input frame image is composed of a third field and a fourth field. When the prediction image of the third field is predicted from the reference image at the first field time, and the prediction of the image of the fourth field is performed from the reference image at the second field time, the time order of the reference image matches the time order of the input image.

Therefore, the field adjustment unit 311
From the field and the second field, a field image having a higher resolution than the original field image is created using the second motion vector. That is, in order to create a first field having a high resolution, an image of a second field that is temporally later is returned to the first field using the second motion vector. If the first motion vector is (Vx, Vy), the second
The field is an image that is one field later than the first field and is １ ／ of the time of the first motion vector.
It is. Therefore, assuming that there is no change in the motion, (Vx /
2, Vy / 2), and this is set as a second motion vector. Therefore, the first field can be temporally matched by returning this portion using the second motion vector (a). In this motion vector
Therefore, returning is the motion compensation prediction based on the motion vector.
The second field when the motion vector is 0 as in the measurement
Compositing while maintaining the pixel position in the frame of
If the torque is not 0, the pixel position where the motion vector is 0
Synthesize by moving the pixel position by the value of the motion vector based on the reference
And creating a high resolution first field.
You.

By creating a new first field from the motion-corrected second field and first field, an image of the first field having the same resolution as the original frame image can be obtained. Conversely, the time of the second field is advanced by using the second motion vector (Vx / 2, Vy / 2) for the first field that is temporally one field later than the second field. (B), and a new second field having the same resolution as the frame is created.

Since the positions of the pixels of these high-resolution field images do not always coincide with the positions of the pixels of the input image, the pixel interpolation unit 312 creates a predicted pixel value by interpolation from pixels near the position of the input pixel. . In particular
As in the conventional motion compensation prediction, the value of the first motion vector is
To make the pixel at the corresponding pixel position the predicted pixel,
If there is no pixel at the pixel position
A predicted pixel is calculated by interpolation from surrounding pixels. This
, The surrounding pixels, ie, the new first field
Of the pixels in the original first field are in the first motion
Pixels around the pixel position indicated by the vector
The pixels in the second field are the first motion vector and the second motion vector.
Image around the pixel position indicated by the
Become prime. As a result, a predicted image can always be obtained from the third field from the first field and the fourth field from the second field. Further, since the resolution of the predicted reference image is the same as the resolution of the original frame, a predicted image with less prediction error can be obtained. This has the effect of increasing the image compression efficiency and improving the image quality of the reproduced image.

[0028]

As described above, according to the present invention,
After field division, each field image created by interpolation or one field image is corrected based on the motion vector and combined with the other field to create a field image corresponding to the field image from the prediction error Therefore, the field of the predicted image does not reverse due to the amount of motion. Therefore, an accurate prediction image can be obtained.

Further, since the matching of the detection of the motion vector is also performed based on the same predicted image, there is an effect that a more accurate and fine motion vector can be detected.

[Brief description of the drawings]

FIG. 1 is a block diagram showing a configuration of a first embodiment of the present invention.

FIG. 2 is a diagram for explaining operations of a field image dividing unit and an oversampling unit in FIG. 1;

FIG. 3 is a block diagram showing a configuration of a second embodiment of the present invention.

FIG. 4 is a diagram for explaining the operation of the field adjustment unit in FIG. 3;

FIG. 5 is a block diagram showing a configuration example of a conventional motion-compensated inter-frame predictive coding apparatus.

[Explanation of symbols]

 101,301 Input terminal 102,109,302,309 Adder 103,303 Orthogonal transformation unit 104,304 Quantization unit 105,305 Multiplexing unit 106,306 Output terminal 107,307 Inverse quantization unit 108,308 Inverse orthogonal transformation Unit 110 Field image division unit 111, 112 Oversampling unit 113, 313 Motion vector detection unit 114 Frame synthesis unit 115, 116, 310 Image memory 311 Field adjustment unit 312 Pixel interpolation unit

 ────────────────────────────────────────────────── ─── Continuation of the front page (56) References JP-A-5-130594 (JP, A) JP-A-6-504169 (JP, A) 1990 Image Coding Symposium (PC SJ90) 175-176 (8-1 Study on adaptive line interpolation inter-field motion compensation method)

Claims (2)

(57) [Claims] 1. A first motion vector generated by using a first motion vector between a reference frame and a frame to be encoded and a first field of the reference frame, and multiplying the first motion vector by a predetermined ratio. A motion compensation prediction method for performing motion compensation on a first field of the encoded frame using a second motion vector and a second field of the reference frame, wherein the predetermined ratio is the first motion vector. A motion compensation prediction method, wherein the time interval is determined by a ratio between the time interval of the second motion vector and the time interval of the second motion vector. 2. A second motion vector generated by using a first motion vector between a reference frame and an encoded frame and a second field of the reference frame, and multiplying the first motion vector by a predetermined ratio. A motion compensation prediction method for performing motion compensation on a second field of the encoded frame using the first motion vector and the first field of the reference frame, wherein the predetermined ratio is the first motion vector. A motion compensation prediction method, wherein the time interval is determined by a ratio between the time interval of the second motion vector and the time interval of the second motion vector.