JPH05199507A - Shift calculation circuit and moving image coder - Google Patents

Abstract

PURPOSE:To generate a motion compensating predictive signal on a frequency area by operating various filters, adding the results of filter calculation and executing shift calculation with precision higher than resolution in a frequency band. CONSTITUTION:This device is provided with filters F11, F21... to operate various filters to respective frequency band signals concerning the signals divided into plural frequency bands and an adder 151 to add the results of the filter calculation at those filters. Namely, since the signals in all the frequency bands from the first to the fourth are added after the filter calculation, the shift calculation of the first frequency band is executed. Thus, position shift calculation is enabled with precision higher than resolution decided at a sample point in the frequency band. Further, totally MXN kinds of filter calculation are executed by respectively executing the N kinds of calculation to the signals divided into the N kinds of frequency bands, the N kinds of filtered results varying the input signals are added, and the shift calculation is executed on the frequency band.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to audio signal and image signal processing, and more particularly to a moving picture coding system.

[0002]

2. Description of the Related Art There are various signal processing methods for dividing a one-dimensional signal or a two-dimensional signal into frequency bands. It is used for compression coding of audio signals and image signals. For this technique, for example, sub-band (subb)
and) division method is adopted as NS Giant and Peter Nor (NS Jayant and Pete).
Prentice Hall, Inc. (PR)
(ENTICE-HALL Inc, USA), AD 198
"Digital Coding of Waveform"("Digital Coding of
Waveforms ") chapter 11 for further details.

This technique is briefly shown in FIG. The input signal is input to the plurality of band pass filters H1, H2, H3, H4 of the sub-band division circuit 60. These filters have different pass band characteristics. After the filtering, appropriate decimation is performed by the decimation circuits 21, 22, 23 and 24 so that the number of samples of the input signal and the number of samples after processing are adjusted to be the same. In the simplest example, N
When using multiple filters, N: 1 decimation is performed after each filter. The above is the sub-band division method, and a method of synthesizing these divided signals is also shown in FIG. Each of the divided signals is interpolated by a zero signal in the interpolating circuits 31, 32, 33, 34 of the sub-band synthesizing circuit 70 to be a signal having the same sample rate as the original signal, and then filtered by the filters G1, G2, G3, G4. Are multiplied by each other and added together by the adder circuit 50.

Several filters have been proposed which guarantee that the signal is completely reconstructed even after the division / combination of them. For example, a quadrature quadrature filter is provided.
reMirror Filter, hereinafter abbreviated as QMF), Conjugated Q
A quadrature filter (hereinafter abbreviated as CQF) and a wavelet filter have been proposed. QM
F is detailed in the above-mentioned "Digital Coding of Waveforms", and CQF is described in, for example, MJT Smith and T. P. Barnwell.
Barnwell) "Exact Reconstruction for Tree Structured Subband Coders", IEE, TransAction on Acoustic Speech and Signal Processing, AD 1986, Vol. 34, p. 434 (Exact reconstruct
tonion for tree-structured
subband coders, "IWEEE Tra
ns. on Acoustic. , Speech and
Signal Processing, Vol. A
SSP-34, pp, 434-441, June 198
6. ) In detail about the wavelet filter, for example, Ingrid Dobby's "Orthonormal Basis of Compactly Supported Wavelets" Communication on Pure and
Applied Massmatics, 1988 AD, 4th
Volume 1, pp. 909-996 (I. Daubechie
s, “Orthonormal Bases of C
ommpactly Supported Wavelet
S., "Commun. on Pure and Ap.
plied Mathematics, Vol. XL
I, 909-996, 1988. ).

[0005]

By the way, since the sample rate is thinned out in each frequency band as compared with the original signal, it may be difficult to operate the original signal easily. For example, consider an example in which the above frequency bands are thinned out to N: 1. In this case, N sample shifts on the original signal correspond to 1 sample shift on each frequency band, and therefore the operation is simple. However, 1 on the original signal
The calculation of the sample shift on the frequency band is not always clear.

However, once the divided signals are combined,
A method is possible in which a shift operation is performed and subband division is performed again. However, this operation involves the operation of sub-band division / combination and becomes expensive in terms of circuit.

It is an object of the present invention to clearly define the shift calculation on the divided frequency bands.

Another object of the present invention is to establish a shift calculation on a frequency band that does not include the operation of subband synthesis / division.

Still another object of the present invention is to provide a motion-compensated interframe predictive coding system on a frequency-divided signal.

[0010]

According to the structure of the first invention, different signals are applied to each frequency band signal with respect to a signal decomposed into a plurality of frequency bands, and the result of the filter calculation is performed. To achieve shift calculation.

Further, according to the structure of the second invention, the shift calculation is performed for all frequency bands.

Further, according to the structure of the third invention, out of the N × N filters used in the structure of the second invention, the filter having a small final contribution ratio does not perform the calculation.

According to the configuration of the fourth invention, the original signal is divided into regions, the shift amount is defined for each region, and the shift calculation realized by the configuration of the first, second, and third inventions is performed for each region. Switching is performed according to the defined shift amount.

According to the structure of the fifth invention, the means for frequency band dividing the two-dimensional signal of each frame with respect to the moving image signal, and the difference between the frequency band divided signal and the prediction signal described later are calculated. Means for obtaining a prediction error signal by calculation, means for quantizing the prediction error signal, means for encoding the quantized prediction error signal, and addition of the quantized prediction error signal and the prediction signal To obtain a locally decoded signal, delay the locally decoded signal by one frame, divide each frame of the moving image signal into regions, obtain a motion vector for each region, and position the motion vector. As a shift amount, the position shift calculation according to the fourth aspect of the present invention is performed on the locally decoded signal delayed by one frame, and means for generating a prediction signal is provided, so that motion compensation on a frequency-divided signal is performed. Realizing predictive coding between frames.

[0015]

As shown in FIG. 7, the interpolation circuits 31, 32, 33 and 34 of the sub-band synthesis circuit 70 interpolate with a zero signal to obtain signals having the same sample rate as the original signal, and then the filters G1, G2 and G3. , G4, and the addition circuit 50 adds them together to restore the original signal expression. Then, the shift circuit 80 performs a shift operation, and again the plurality of band pass filters H1, H of the sub-band division circuit 60.
2, H3, H4 and after filtering, thinning circuit 2
The operation when subband division is performed by appropriately thinning out 1, 22, 23, and 24 is formulated. First, sub-band synthesis, shift, and sub-band division are

[0016]

[Equation 1]

It can be expressed as Where b _l (n)
Is a signal of each frequency band, g _l (n) is a filter for the l-th band used for subband synthesis, x (n) is a synthesized signal, y (n) is a shifted signal thereof, and r is a shifted signal The quantity, h _l (n) is the filter for the l-th band used for sub-band division, and b _l ′ (n) is the shifted signal of each frequency band. Combining these three equations,

[0018]

[Equation 2]

From this equation, it is understood that the shift calculation on the frequency band can be realized by performing filter calculation on the signals in each frequency band and adding them together. This proves the validity of the configuration of the first invention.

In this configuration, the shift calculation is applied to only one frequency band, but it can be applied to all frequency bands. This is the configuration of the second invention.

In the configuration of the second aspect of the invention, since N kinds of filters act on the signals divided into N as shown in FIG. 2, a total of N × N kinds of filters are required. That is, it is considered that the correction term is added from another frequency band in addition to the filter calculation in the frequency band.

If the frequency band is ideally divided, these correction terms can in principle be eliminated. However, a filter that realizes an ideal frequency band division is not possible with a filter having a finite tap, and a filter having a finite length must be used, and such a correction term is necessary. It is conceivable that such a correction term has a significant value in adjacent frequency bands, but has a very small value between distant frequency bands. Therefore, in the third invention, N ×
Instead of using all N types of filters, the filter calculation for generating the correction term with such a small contribution is not performed.

Further, in the fourth invention, the signal is divided into regions,
The shift amount is switched for each area.

The fifth and final invention is a structural example applied to a moving picture coding system. As shown in FIG. 5, the input signal is subband-divided in the subband division circuit 60, the subtractor 202 takes the difference from the prediction signal 300, and the quantizer 2
Quantize with 02 and encode. The quantized signal is
The inverse quantizer 204 restores the original signal, the adder 205 adds the prediction signal 300, and the frame delay unit 206
Then, the frame is delayed by 1 frame and is subjected to motion compensation prediction by the shift calculation circuit 200 to become a prediction signal 300. This motion compensation portion is a shift calculation using the motion vector for each sub-block area, which is separately obtained by the motion amount detector 207, and the above-mentioned shift amount is shifted after the frequency band division is switched for each area. Calculation method is applied.

[0025]

DESCRIPTION OF THE PREFERRED EMBODIMENTS FIG. 1 shows a shift calculation circuit according to the structure of the first invention. This shift calculation circuit applies filters F11, F21, F that apply different filters to each frequency band signal with respect to a signal divided into a plurality of frequency bands.
31 and F41, and an adder 151 for adding the results of those filter calculations. The shift calculation of the first frequency band is performed by performing filter calculation and addition of the signals of all the first to fourth frequency bands. With this circuit, it is possible to realize the position shift calculation with an accuracy equal to or higher than the resolution determined by the sample points in the frequency band.

FIG. 2 shows a shift calculation circuit according to the structure of the second invention. This shift calculation circuit has a signal processing circuit that realizes in-frequency band position shift calculation for all frequency bands of a signal decomposed into a plurality of frequency bands. The signal processing circuit includes filters F11, F21, which apply different filters to the respective frequency band signals.
F31, F41, an adder 151 for adding the results of the filter calculations, and filters F12, F22, F3 for applying different filters to each frequency band signal.
2, F42 and an adder 152 for adding the results of the filter calculations, and filters F13, F23, F33, F for applying different filters to each frequency band signal.
43 and an adder 153 for adding the results of those filter calculations, a filter F14, F24, F34, F44 for applying a different filter to each frequency band signal, and an adder 154 for adding the results of those filter calculations. is doing. According to this shift calculation circuit,
The shift calculation is performed not only on the first frequency band but also on signals in all frequency bands.

FIG. 3 shows a shift calculation circuit according to the structure of the third invention. This shift calculation circuit has a filter F1.
1, F21 and an adder 151 for adding the results of the filter calculations, and filters F12, F22, F3
2 and an adder 152 for adding the results of the filter calculations, and an adder 15 for adding the filters F23, F33, F43 and the results of the filter calculations.
3 and a filter F34, F44 and an adder 154 for adding the results of those filter calculations. Compared with the configuration of the shift calculation circuit of FIG. 2, a filter from the first frequency band to the third and fourth frequency bands, a filter from the second frequency band to the fourth frequency band, and a third filter The filter from the frequency band to the first frequency band and the filter from the fourth frequency band to the first and second frequency bands are omitted.

FIG. 4 shows a shift calculation circuit according to the structure of the fourth invention. In addition to the configuration shown in FIG. 3, this shift calculation circuit supplies a signal indicating the shift amount to the filters, and each filter switches the characteristics according to the shift amount.

FIG. 5 shows a moving picture coding apparatus using a shift calculation circuit according to the structure of the fifth invention. This moving image coding apparatus calculates a difference between a frequency band-divided signal and a prediction signal by calculating a difference between a frequency band-divided signal and a prediction signal for a moving image signal, and a sub-band division circuit 60 that divides the two-dimensional signal of each frame into frequency bands. , A quantizer 203 that quantizes the prediction error signal, an inverse quantizer 204 that encodes the quantized prediction error signal, and a quantized prediction error signal and the prediction signal. An adder 205 that additionally obtains a locally decoded signal, a frame delay unit 206 that delays the locally decoded signal by one frame, each frame of a moving image signal is divided into regions, and a motion amount that obtains a motion vector for each region A detector 207 and a system for generating a prediction signal by performing the position shift calculation of the fourth invention on the locally decoded signal delayed by one frame by using the motion vector thereof as a position shift amount. And a preparative calculation circuit 200.

In this moving picture coding apparatus, the input signal is divided by the sub-band dividing circuit 60, and the subtracting circuit 202.
, The difference from the prediction signal 300 is calculated, and the quantizer 203
Is quantized and encoded by. The quantized signal is returned to the original by the inverse quantizer 204, the prediction signal is added together by the adder 205, delayed by one frame by the frame delay unit 206, and subjected to motion compensation prediction by the shift calculation circuit 200. It becomes the prediction signal 300. This motion compensation part is a shift calculation using the motion vector for each sub-block area, which is separately obtained by the motion amount detector 207, and the shift calculation is performed after the above-mentioned shift amount is divided into frequency bands. Method is applied.

[0031]

According to the shift calculation circuit of the present invention, the shift calculation can be realized while being divided into frequency bands. In particular, it is possible to generate a motion compensation prediction signal in the frequency domain in a moving picture coding device.

[Brief description of drawings]

FIG. 1 is a shift calculation circuit according to a configuration of a first invention.

FIG. 2 is a shift calculation circuit according to a configuration of a second invention.

FIG. 3 is a shift calculation circuit according to a configuration of a third invention.

FIG. 4 is a shift calculation circuit according to a configuration of a fourth invention.

FIG. 5 is a moving picture coding apparatus using a shift calculation circuit according to the configuration of the fifth invention.

FIG. 6 is a diagram illustrating subband division / synthesis.

FIG. 7 is a diagram illustrating subband synthesis, shift, and subband division.

[Explanation of symbols]

21,22,23,23 4: 1 decimation circuit 31,32,33,33 1: 4 interpolation (interpolates zero signal) circuit 60 sub-band division circuit 70 sub-band synthesis circuit 80 shift circuit 151,152,153, 154 adder 202 subtractor 203 quantizer 204 inverse quantizer 205 adder 206 frame delay 207 motion amount detector H band pass filter used for subband division G bandpass filter used for subband synthesis F shift calculation The filter used to do

Claims (5)

[Claims] Claims: 1. For a signal decomposed into a plurality of frequency bands, means for applying different filters to each frequency band signal, and means for adding the results of the filter calculations are included. A shift calculation circuit that realizes a position shift calculation with a resolution equal to or higher than the resolution determined by the sample points in. 2. A shift having a signal processing circuit for realizing the position shift calculation within the frequency band according to claim 1 for all the frequency bands with respect to a signal decomposed into a plurality of frequency bands. Calculation circuit. 3. A shift calculation circuit characterized in that, when the position shift calculation in the frequency band according to claim 1 is realized, a filter calculation having a small contribution of the position shift calculation is omitted. 4. A means for performing the position shift calculation according to claim 1, 2 or 3 for a one-dimensional or two-dimensional signal which is divided into regions and in which the amount of position shift is defined for each region. A one-dimensional or two-dimensional shift calculation circuit having means for adaptively switching the position shift calculation for each area. 5. A means for dividing a two-dimensional signal of each frame into frequency bands with respect to a moving image signal, and a means for obtaining a prediction error signal by calculating a difference between the frequency band divided signal and a prediction signal described later. A means for quantizing the prediction error signal, a means for coding the quantized prediction error signal, and a means for adding the quantized prediction error signal and the prediction signal to obtain a locally decoded signal 5. The position according to claim 4, wherein the means for delaying the locally decoded signal by one frame, the means for dividing each frame of the moving image signal into regions, and the motion vector for each region, and the motion vector as a position shift amount. And a unit for performing a shift calculation on the locally decoded signal delayed by one frame to generate a prediction signal.