JPS6028392A - Movement interpolation system of dynamic picture signal - Google Patents

Abstract

PURPOSE:To obtain a movement interpolation circuit by giving an identical representing dynamic vector to a dynamic area corresponding to a moving object and applying the moving interpolation based thereupon so as to eliminate the discontinuity of an interpolated picture. CONSTITUTION:An input picture signal is given to a delay circuit 10 and an optimum forecast discriminating circuit 11. The optimum forecast discriminating circuit 11 is a dynamic vector detection circuit and seeks the optimum forecasting system. The delay circuit 10 delays an input signal for a time's share detecting an optimum forecasting system and gives the result to a subtraction circuit 12. A variable delay circuit 13 gives an optimum forecasting value to the subtraction circuit 12 based on the optimum forecasting system. The subtraction circuit 12 generates a forecast error signal and outputs it to a quantizing device 14. The quantizing device 14 quantizes the forecast error signal based on quantized control information and outputs it. An adder circuit adds the optimum forecast signal and the quantized forecast error signal, generates a local decoding signal and outputs the signal to a frame memory 15. The frame memory 15 delays the local decoding signal for nearly one frame time's share and outputs it to a variable delay circuit 13. The output of the quantizing device is fed to a compression coding circuit 17 and the output is compressed by using the method of inequivalance coding.

Description

DETAILED DESCRIPTION OF THE INVENTION The present invention relates to an interpolation method for TenVision signals using motion vectors.

Conventionally, several methods are known as interpolation methods for TenVision signals using motion vectors.

To simplify the explanation that follows, three temporally consecutive frames'! Let i7A, B%C, and let B be the thinned out frame, the frame before A'iB, and C the frame after B. For example, in one method, a motion vector for each block between ACs is calculated, and based on this motion vector,
For each block in frame B, a motion vector between AB is determined 41L. Next, starting from the position of the block to be interpolated in frame B, the block corresponds to the position specified by the inverse vector that has the same magnitude as the motion vector estimated for that block and can be reversed. is placed in frame A, and the image signal of this block is used as the image signal of the block in frame B to be interpolated. By performing this operation for all the pixels in frame B, the interpolated image signal for frame B is obtained.

In another method, the motion vector is determined for each Glock between frames A and C, and based on this,
A motion vector between AB is estimated, and this estimated motion vector is assigned to each block (first block) in frame A. Next, starting from these first blocks, a second glock corresponding to the position specified by the assigned motion vector is placed in frame B, and with the image signal of the first glock in frame A, Let it be the image signal of the second block in frame B that is desired to be interpolated.

By performing this operation on all the clocks in frame A, the interpolation signal for frame B is obtained.

The above two methods are examples, but in both of these methods, the interpolated image signal is obtained in units of blocks divided into a certain size within 7somes, resulting in motion vector detection errors. In such cases, even within a moving region corresponding to a single moving object, spatial discontinuities occur in units of 7 degrees, resulting in very noticeable image quality deterioration such as disconnection of continuous lines. There were drawbacks to the interpolation plane J image.

An object of the present invention is to eliminate discontinuity in interpolated images with respect to moving regions, and to provide better motion interpolated images.

That is, according to the present invention, the frames A and C before and after the thinned out frame B are processed in units of 2. ℃・ is the motion vector determined for each pixel and the motion region, and from the motion vector determined for each Glock or pixel within each motion region, one motion vector representative of that motion region is calculated. The representative motion vector is calculated and assigned to that motion area.
For each moving region, a motion-inside method for a moving image signal is obtained in which the image signal of the moving region can be obtained by interpolation from one or both of frames A and C.

The principle of the present invention will be explained below.

First, several methods are already known for calculating a motion vector. For example, 1. In a method called the munching method, the flame-sensitivity signal before the frame of interest is shifted based on the same position of the frame of interest (the vector indicating the amount of shift and the direction of shift is called a shift vector). , p7 and the evaluation function determined from this difference signal (for example, 2 of the difference signal
The shift vector that minimizes the value of the evaluation function is determined as the motion vector of the Glock. In a method called the gradient method, a motion vector is calculated from the gradient of the luminance signal within a frame and the difference signal between the previous and subsequent frames. You can determine the motion vector between you and this person. Methods for determining a moving area include a method in which the moving vector determined by the above method is a Glock that does not create an atmosphere, or a method in which a set of pixels is used as a moving area, and a method in which the moving area is determined by the difference in brightness level between frames, that is, the absolute value of the frame difference signal. A method can be used in which a set of pixels whose values are larger than a certain threshold is defined as a moving region.

In the present invention, any method of detecting a motion vector or determining a motion region may be used. Next, a method for calculating a motion vector representative of a motion area will be described using an example in which a motion vector is detected in units of Glocks.

First, a function indicating a moving area is defined by the following equation, where (XllX2) is a coordinate representing the position of a block within a frame.

Next, similarly, let the motion vector detected for Glock (X1lx2') be V (Xl, X2) and M (X
A new motion vector sequence is calculated from the product of V (Xl, X2).

';' (XllX2) = M (xrrxx)
・Since the y(X8.x2) motion vector is a two-dimensional vector, for V'(x+ + Alternatively, the average vector Va of φ(X++ The interpolation motion vector VM (X11
X2 ), and starting from the position of the block where you want to obtain the interpolated image signal, for example, 1 of VM (x, , x, )
/2 inverse vector, -1/2 VM (X++Xz)
ic L-). The image signal of the block in frame A is taken as the image signal of the block. Alternatively, starting from the position of a block in the motion area in frame A, and using the image signal of this block, 1/2 VM (X I+X
z). This is the image signal of the block (the block concerned). This operation is performed on a block-by-block basis, but since the same representative motion vector is given to all blocks within the motion region, motion interpolation is performed using each motion region as one unit. This means that In fact, in the case of a short time interval such as the frame time, the moving object can be approximated to the motion of a rigid body without any change in shape, and there are fewer errors in detecting the motion vector. The rigid body 41 constant interpolation method that uses a moving region as one unit can provide a natural interpolated image. Further comparisons between the conventional motion interpolation method explained above and the motion interpolation method according to the present invention are shown in FIGS.
Add an explanation using B. In FIG. 1, for example, a symbol "1" moves between frames and is divided into blocks (in this case, divided into seven), which are displayed with arrows. When looking at the frequency distribution of the motion vectors detected from FIG. 1, 1 is the most frequently detected four representative motion vectors. Figure 2A corresponds to the case where an interpolated image is created independently using the motion vectors for each professional, as in the conventional method, and the pattern of ``1'', which is originally continuous, is an error in the detection of the excitation vector. Therefore, it is one discontinuous interpolated image.

Figure 2B corresponds to the case where the representative motion vector shown in Figure 1 is interpolated with the symbol "1" as one motion area, and a spatially continuous interpolation image is created. Obtainable.

Next, embodiments of the present invention will be described in detail with reference to the drawings. FIG. 3 shows an embodiment on the transmitting side. The input image signal is Kizuna 1010. A! The signal is sent to the delay circuit 10 and the optimal prediction determination circuit 11 via 111o+1. The optimal prediction determination circuit 11 is also a motion vector detection circuit, and a motion vector is detected using, for example, the method described in the above-mentioned matching method, and searches for the optimal prediction method that provides the minimum prediction error or the minimum amount of code. The delay circuit 1o has an optimal predetermined value 11.
Delay the great sword signal by the time it takes for 1.11 million types to be detected,
via line 1012 to subtraction circuit 12. The variable delay circuit 13 sends the optimal predicted value to the subtraction circuit 12 through the gland 1312 based on the optimal prediction method obtained through the 'K1113'.
supply to The subtraction circuit 12 generates a difference signal between the delayed long sword signal and the optimal prediction l1llL, that is, a pre-611j error signal, and outputs it to the quantizer 14. Quantizer 1
4, the prediction error signal is quantized and output based on the ``proponentization control information'' supplied via *J 1814,
If the number corresponds to seven atoms to be thinned out, the output level is quantized over one frame time, for example. In the adder circuit 16, the signal supplied via line 1316 is the same as the signal output to 1312), i.e. the optimal preff111 signal;
The sum of the prediction error signals supplied via the line 1416 is summed to generate a local signal, which is output to the frame memory 15. The frame memory 15 is capable of storing approximately one frame of the input image signal, and delays the locally decoded signal by approximately 17 V-times 1 d.
The signal is output to the variable delay circuit 13 via. Quantizer 14
The output of the quantized prediction error signal is sent to the compression encoding circuit 17 and compressed by using the unequal length encoding method.
d, but if it corresponds to a thinned out frame,
The data is outputted to the buffer memory 18 with the thinning information supplied via the buffer memory 17 added thereto. Furthermore, in the case of the above-mentioned matching method, information representing the optimal prediction method supplied via line 1117 is also added. The buffer memory 18 outputs these compressed and encoded signals to the transmission path 2000 so that the transmission rate is constant, and also decimates the quantization control information to the quantizer 14 via the wire 1814 via the wire 11817. The information is supplied to the compression encoding circuit 17. FIG. 4 shows an embodiment on the receiving side.

The compression-encoded prediction error signal, frame thinning information, and information representing the optimal prediction method (motion vector information) supplied via the transmission path 2000 are decoded and expanded by the expansion circuit 20 . The line 20'Z9 supplies the decoded frame thinning information to the interpolation circuit 25, and the line 2024 supplies the frame thinning information and the optimal prediction method to the motion vector decoding circuit 24. Further, the prediction error signal is output to the adder circuit 21 via the heel 2021. The adder circuit 21 adds this prediction error signal and the optimal prediction signal inputted from the variable delay circuit 23 via a line 2321 to create a new decoded image signal, and sends it to the interpolation circuit 25 via a line 2125. To, m212
2 to the frame memory 22. The frame memory 22 stores the decoded image signal for approximately one frame, and the frame memory 22 stores the decoded image signal for approximately one frame.
3 to the interpolation circuit 25 via line 2225. The variable delay circuit 23 variably delays the decoded image signal output from the frame memory 22 based on the optimal prediction method supplied via the line 2423, and outputs it to the adder circuit 21 via the line 2321. The motion vector decoding circuit 24
The optimal prediction method (motion vector) is determined from the frame thinning information supplied via 024 and the information representing the optimal prediction method, and is output to the interpolation circuit 25 via line 2425 as well as via line 2423. , to the variable delay circuit 23, a zero vector is output in the frame thinning state, and the above-mentioned optimal prediction method is output in other states. The interpolation circuit 25 performs motion interpolation on the thinned out frames using the decoded image signal, the optimal prediction method, and the frame thinning information, and outputs the motion interpolation to the frame 3000 in a regular time order.

The interpolation circuit 25 will be explained using FIG.
125 and delays the decoded picture signal supplied via line 3.
This is output to the selector 37 via 037. The motion vector memory 31 stores about one frame worth of motion vectors supplied as the optimal prediction method via the gland 2425.
I%! 3132 to the frequency distribution calculation circuit 32 and t1
It is output to the motion area determination circuit 34 via M3134. The frequency distribution calculation circuit 32 includes a counter for the number of vectors within the motion vector detection range, an X component of the motion vector,
It consists of a part that specifies a counter to be used for counting according to the Y component, and calculates the distribution of frequency of occurrence for l frames of motion vectors and outputs it to the comparator 33. Comparator 3
3 compares the frequency of occurrence of motion vectors for one frame, finds the maximum motion vector, and draws this as line 333.
5 to the multiplier 35. Dynamic area determination circuit 34
consists of a motion vector norm calculation unit, a comparator, and a motion region memory, and calculates the norm of the motion vector supplied via the line 3134, and compares the magnitude of the norm with a predetermined threshold. , if the norm of the motion vector is larger, it is determined that it is a motion region and outputs 1, otherwise it is determined that it is a stationary region and outputs O to the motion region memory, and this is transferred to line 3.
435 to the multiplier 35. The multiplier 35 multiplies the motion vector with the highest frequency of occurrence (referred to as the representative vector) supplied via the line 3335 by the motion area information (1 or 0) supplied via the line 3435. death,
That is, all representative motion vectors are given to the dynamic region, and zero vectors are given to the static region, and outputted to the variable delay circuit 36 via m''' 3536. Based on the motion vector (representative motion vector or zero vector) supplied through
With a variable delay, the decoded image signal supplied via the line 36
The signal is outputted to the selector 37 as an interpolation signal via the signal line 37.

The selector 37 selects the line 30 if it is not in the thinning state according to the frame thinning information supplied via m2025.
When the decoded image signal supplied via line 3637 is thinned out, the interpolation signal supplied via line 3637 is selected and output.

As explained in detail above, according to the present invention, the same representative motion vector is given to the motion area corresponding to the moving object and motion interpolation is performed using this representative motion vector.
Even if some motion vectors are detected incorrectly, the interpolated images are continuous and it is possible to obtain better image quality.

[Brief explanation of drawings]

Figure 1 is a diagram showing an example of a target image and a detected motion vector example, Figure 2A is a diagram showing an example of an interpolated image without the rigid body assumption, and Figure 2B is an example of an interpolated image with the rigid body assumption. A diagram showing
FIG. 3 is a diagram showing an example of a transmitting circuit used in implementing the method of the present invention, FIG. 4 is a block diagram showing an embodiment of the present invention, and FIG. 5 is an interpolation circuit shown in FIG. 4. Detailed circuit ψ1
] is a figure showing. In the figure, 10. Delay circuit, 11. Optimal child ω1] published circuit,
12. Subtraction circuit, 13. Variable delay circuit, 14. Quantizer, 15・Frame memory, 16. Addition circuit, 17. Compression encoding circuit, 18. Buffer memory, 20. extension circuit,
21. Addition circuit, 22. Frame memo IJ, 23. variable delay circuit, 24. Motion vector decoding circuit, 25. Motion interpolation circuit, 30. Delay circuit, 31. Motion vector memory, 32
, frequency distribution calculation circuit, 33° comparator, 34. Dynamic area determination circuit, 359 multiplier, 36. Variable delay circuit, O 1 Fig. 2? 2 Fig. A Piece 2 Fig. B 1-7I-3 Fig.... 11t O 4 Fig. O 5 Fig.

Claims (1)

[Claims] In a method of interpolating the screen of a thinned out frame from the screen before and after the thinned out screen for a moving image signal consisting of screens in frames or field units, Motion vectors representing each of the motion regions that are considered to be one or different motions in the screen are calculated, and each motion region is treated as one unit by these representative motion vectors, and the thinned out frame is A motion interpolation method for moving image signals in which vfa is used to obtain an image signal for each moving region in the thinned out frame by interpolation from 10,000 frames before and after , or both.