JPS6295084A - Highly efficient encoding device for television signal - Google Patents

Abstract

PURPOSE:To stably identify a parameter by using a data by obtained by superimposing a random noise on the picture element data of a past field as a reference data. CONSTITUTION:To a parametr identification part 1, the picture data of a present field (k) is inputted and the reference data of the past three fields are supplied from respective adding circuits 6, 7, 8. To the adding circuits 6, 7, 8, white noises from a white noise generating circuit 9 are commonly supplied, and to the adding circuit 6, the picture data of a preceding field (k-1) is supplied. To the adding circuit 7, the picture data of a further preceding field (k-2) is supplied. To the adding circuit 8, the picture data of still further preceding field (k-3) is supplied. The picture data of these part fields are the prediction data and by using the picture data obtained by superimposing the white noise on this prediction data and the data of the present field, the parameter identification pat 1 identifies 35 parameters having 8 bits at every parameter w1-W35, for instance, at every one field by a least square method.

Description

DETAILED DESCRIPTION OF THE INVENTION [Field of Industrial Application] The present invention relates to a high-efficiency encoding device for television signals.

[Il'r Construction of the Invention:- This invention is based on a high-efficiency coding method that reduces the number of vias per pixel in digital television, and converts pixel data from several past fields into the current field. When predictively encoding pixel data of This makes it possible to stably obtain .

(Prior Art) An interframe encoding method that performs three-dimensional, ie, spatiotemporal processing is known as a high-performance encoding method that reduces the number of bins per pixel.Interframe encoding method There are two methods: one based on motion detection and the other based on motion compensation.The former detects motion based on whether or not there is a frame difference, and uses +f as the data of the previous frame where there is no frame difference (that is, where there is no movement). It is something that can be called out.

The latter method uses a block matching method to obtain positional relationship information (motion correction amount) between the current and previous frames, and then manipulates the previous frame image based on this motion correction amount to create correspondence between frames. The block matching method divides the screen into a plurality of blocks, determines the amount of movement and its direction for each block, and transmits the amount and direction of movement.

[Problem that the invention seeks to solve]

The interframe coding method using motion detection has the problem that general moving images have many moving parts and a low compression rate.

Furthermore, the interframe coding method using motion compensation has the disadvantage that the compression ratio is not sufficiently low because distortion occurs due to block division and the amount of motion is transmitted for each block.

Furthermore, both methods have the drawback of causing a so-called uncovered background problem in which pixel data in the original area disappears when a moving object moves.

Therefore, it is an object of the present invention to provide a highly efficient encoding device that can achieve an extremely high compression ratio compared to conventional devices.

Another object of the present invention is to provide a highly efficient encoding device that can cope with various movements caused by a plurality of moving objects by performing various corrections in the time direction.

Still another object of the present invention is to provide a highly efficient encoding device that eliminates problems such as edge blurring and anchor-hard grout by performing various corrections in the spatial direction.

In addition, the present IN applicant previously filed a patent application No. 59-1 for a highly efficient encoding device for television signals that can achieve an extremely high compression rate.
74412). The object of the present invention is to improve this high-efficiency encoding device.

In other words, what is shown in 1+i above is to obtain a predicted value for a pixel in the current field by extracting a nearby pixel with the strongest correlation as a representative value, and applying correction in the spatiotemporal direction to this representative value. The parameters for
! III is identified to minimize the sum of squared errors.

In calculating the parameters that minimize this sum of squares, it is necessary to solve the inverse matrix.

However, due to limitations on the word length of still images, there was a problem in that the conditions for finding the inverse matrix were not satisfied in video scenes with almost no movement that could not be distinguished from still images, resulting in indeterminacy.

Therefore, an object of the present invention is to provide a highly efficient encoding device for television signals that can stably obtain a harameter.

Solution to Problem C] This invention uses a memo 1) for storing pixel data of several past fields, pixel data of the current field, and pixel data of several past fields stored in memory. Means 1 identifies parameters that define a spatio-temporal relational expression defined by linear precision from pixel data on which random noise is superimposed, for example, by the method of least squares, and based on the identified parameters, and means for predicting pixel data of a current field from pixel data of several fields, and transmits the identified parameters.

[Effect]

This invention predicts current motion from pixel data of several past fields. In this invention, since the motion information (υ) of each of a plurality of moving objects is included in the above pixel data, in other words, motion vectors with various directions and speeds also have a strong temporal correlation, There is no need to transmit the amount of motion correction, and only the parameters (coefficients for prediction) for each field need to be transmitted, and the average number of bits per pixel can be extremely reduced. is captured as a temporal change in the level of each pixel, so it is a constant velocity VRG motion that is independent of the direction and speed of the motion vector (represented by the data of the past two fields).
! ! (
Therefore, it is only necessary to correct the deviation from the motion model in a single step. Therefore, according to the present invention, the compression ratio can be increased. Furthermore, since the correction is performed three-dimensionally (temporally and spatially), problems such as block distortion and uncovered hackground do not occur. Furthermore, in this invention, since the reference data for parameter identification is superimposed with random noise, it is difficult to use static images or images with almost no movement.
Parameters can be stably identified even in the case of an elephant.

[Embodiment 1 G1] Hereinafter, an embodiment of the present invention will be described with reference to the drawings. A description of this embodiment is given below in section 111a.

a. Encoding device C. Decoding device C. Identification of parameters a. Encoding device FIG. 1 shows an embodiment 1 of the present invention. II send fullll
The configuration of an encoding device is shown below.

In FIG. 1, 8 and 1 indicate a parameter identification section. A digital television signal digitized at a predetermined sampling frequency, that is, image data of the current field, is input to the parameter identification section 1, and reference data of three past fields are input to the addition circuit 6.7. .8. The adder circuits 6, 7, and 8 are commonly supplied with white noise from the white noise generation circuit 9, and the power n calculation circuit 6 is supplied with the image data of -1 in the previous field. Circuit 6 is -2 in the previous field.
image data is supplied. The image data of -3 is supplied to the power n calculation circuit 8 in the field further before.

The image data of these past fields is 1,71) 1 data, and using the image data obtained by superimposing white noise on this predicted data and the current field data, the parameter identification unit 1 calculates, for example, 8 bits each using the least multiplicative method. Thirty-five parameters W1 to W3.4 are identified for each l field. The parameter identification section 1 includes a line connection circuit and a sample delay circuit for adjusting spatial positional relationships. Parameter W identified by parameter identification unit 1
1 to w35 are sent data. This parameter W1
~w 35 is 1 field behind the input data 1
This field is -1.

2 indicates a prediction unit, and 3.4.5 indicates a field memory, respectively. The field memory 3.4.5 is loaded with the prediction data from the prediction unit 2, and the image data of -1°k-2, -3 in the past three fields from -1 in the field (pre-prediction 1 data) can be stored. The prediction unit 2 uses 35 (cause prediction data and parameter W1) located in the vicinity of the pixel to be predicted and included in the past three fields.
~ w35 is used to obtain a predicted value for the current pixel. For this reason, the prediction unit 2 also includes a plurality of line delay circuits and a plurality of sample delay circuits for adjusting the spatial positional relationship.

For the pixel data (Fig. 3A) of the current field) Sl
0 pixel data (Figure 3B), in the field two before that -
Data of 15 pixels near 2 (Fig. 3C), data of 10 pixels near -3 (3rd
It is obtained as a linear combination of a total of 351 pixel data (white noise is superimposed) in Figure D).

In Figures 3A to 3, the solid horizontal lines represent the lines scanned in field k and field F'k24, and the dashed horizontal lines represent the lines scanned in field -1 and field -3. .

represents a line. The line at the position where the pixel data in the current field is included is designated as y, the line located above it is designated as y+1, and the line located on the -F side of the historical input y1-1 is designated as y+2. The lines located below line y are designated as y-1 and y-2, respectively.

Figure 3A - In the third diagram, the vertical solid line indicates the sampling position in each field, and the sampling position 1 sample before the sampling position X of pixel data in the current field and the sampling position 2 from this 3t. The sampling positions before the sample are x-1 and x-2, respectively. Further, the sampling position i9 after the sampling position X and the sampling position after the sampling position are x+l and X-1-2, respectively.

The predicted data ↑b (x, y) with respect to the current pixel is expressed by the + star-order combination of the following equation. 8.
The pixel data of the past fields 1, . . . are those on which white noise is superimposed.

↑ y (x, y) =
wl x ↑ k-+ (x-2, y+1)
−w2×↑−I(x−1,yll) I−w3 x
↑, −, (x, y+1)”w4 × im−+
(x+Lytl) 4- w5 x 'L
-+ (x+2.y+1)-W6 × end-+
(x-2,y-1) -i-w7 -1(x
-1, y-1) = w3 X fk-+ (x, y-1)
+ w9 X 7 h-+ (x+1.y-1)"
wlQX f k-+ (x'r2.v-1)+-w
ll×↑h−z (x−2, y□2) “w12X
Th-z (x-Lv+2)'-w13X 'i'-
2(x, y+2) 'w14X: completeh-z (x+
Ly+2)=w15X Tk-z(x;2.y+
2) −w15X↑−z (x−2,y)” w1
7x T k-z (x-1,y) -W18X T
k−z (x, y)−=w19X了b−z (x4
Ly) ′-W20X↑w-z (x+2.yJ+
w21X-2 (x-2, y-2) t W 22
-2 (x-1, y-2) + w23X ↑v-z to X block
(X, y-2) '-w24x Tk-z (
x+1. y-2)! W25X T k-z (x+
2,y-2)↓W26×↑w-3(X-2+ν4)
+ w27 X T k, + (x4.y±1)+
w28x↑y-* (x, yll) + w29X
↑-t (x*1+'y+4)" w3QX T
h-* (x+2.c1) t w31 X T
k-:t (x-2, y-1)+W32X completed h-y
(X-1, y-1) ↓W33×↑,,□1(
x, y-1)t w34X T k-1(x*Ly4)
+w3;)X↑w-z (x+2.y-1)-)
- The above prediction formula means that the predicted value for the pixel in the current field is obtained by extracting the neighboring pixel with the strongest correlation as a representative value, and applying correction in the spatiotemporal direction to this representative value. .

The parameter identification unit 1 identifies parameters by the least squares method using the reference data from the addition circuit 6.7.8. In other words, the true value Ik of the pixel in the current field is -
Since the error e is superimposed on the predicted value ↑ of the pixel corresponding to the hole obtained using the L formula, (e−↑v I
+=), and parameters W1 to w35 that minimize the square of this error for a predetermined number of pixels are calculated.

In case 2, all predicted pixels (n) included in one field (
For example, 800 pixels in l line, 1 field is 255
In the case of a line, the highest accuracy can be obtained by calculating the parameters W1 to w35 by the least squares method using 800x2551171I). For example, using 300 representative pixels, the power meter W1~w
35 identifications are made.

Also, in areas where there is no data at the periphery of the screen, the fourth
As shown in the figure, the same data as 1, data on the screen, ] to h can be substituted for data outside the screen. Alternatively, as shown by the broken line in FIG. 4, identification may be performed within a region that is one line inside and two samples inside.

Note that it is also possible to use pixel data of 2 fields)', which is too large for the current field, and in that case, 3-dimensional 11
i1! As a dynamic model, a constant velocity motion model will be expressed.

Further, instead of white noise, random numbers at a small level of about 0 to 4 may be used.

b. Decoding device The decoding device that receives the encoded transmission data described above has a field memory 12.1 as shown in FIG.
3.14, and a prediction unit 11 which is supplied with the received parameters W1 to W35 and is supplied with data of the past three fields from the field memory 12.13.14.

This t-measuring section 11 forms restored data, that is, a digital television signal. On the receiving side, parameters W1 to W35 are used to restore the digital television signal.
Prior to the transmission of , the initial values for 3 fields are transmitted,
This initial value is written into each of the field memories 12, 13, and 14.

C. Identification of Parameters An example of the identification of the parameters W1 to w35 performed by the parameter identification unit 1B described above using the method of least squares will be described below.

The linear one-component combination equation for calculating the prediction data ↑y (x, y) described above can be expressed by the following determinant when prediction is performed for the entire current field.

1 ゛If the above equation is expressed collectively using a matrix and a vector, T
The matrix W of n, 35) is a 35th order vector.

On the other hand, the vector I formed by arranging the data (page values) of the current field is (m
Xn) Next pre-ill 2% difference vector, then -4+e=↑middleW+e. The above formula becomes e-cho1cho・W. A parameter ca that minimizes the square of this prediction error vector e is determined. The above formula is transformed as follows. Then, T indicates a transposed matrix.

e" e=CI-TW Bi(■-↑W) =zT l-1T TW-W," ↑1 ■Book W"T
"TW in the above equation, e" The parameter W that minimizes e satisfies the following equation. The derivation of this formula can be found, for example, in Reference 1, System Identification 5 (Publisher 2, Society of Instrument and Control Engineers 1, Publication date: February 10, 1980 (first edition)).
It is described in Chapter 4, Section 4, Item 2.

w ,', w=(↑T T )−↑1■ According to the above, in order to stably find the parameter W,
)−' must be stably determined. In this embodiment, even in the case of a static W image or a moving image with almost no movement, white noise is superimposed, so the inverse matrix can be stably determined.

If this continues, in the case of all (mxn) pixels in one field, a very large matrix of (y)1xn, 25) will be handled, which is not practical. Therefore, the above equations are processed by converting them into matrices and vectors of small order. That is, the (35, 35) matrix of (P=↑1 ・↑),
(Q-↑1 ・D)'s 35th order Hek I ・ Le is used.

x l Tk-, (x, -2+y; +1) ↑h-
+ (Xl-1+y, +l)] ・ ・ ・ T k-:I (Xi 1,000 2.yj-
1) Knee”X? y (x, +y=) The above-mentioned P and Q are formed from the past three fields of prediction data supplied to the balun identification unit 1. and,
Parameter W is calculated by (P-'Q).

〔Effect of the invention〕

This invention predicts the current motion from the past several fields of riIII elementary data. Therefore, since the motion information of each of a plurality of moving objects is included in the above pixel data, the amount of motion is transmitted. There is no need to do this, and it is sufficient to transmit only the parameters (coefficients for prediction) for each field. (2) The average bits E1 per pixel can be extremely reduced.

In addition, in this invention, since motion correction is regarded as a temporal change in the level of each pixel, constant velocity motion (expressed by data of the past two fields) or constant acceleration motion (expressed by data of the past two fields) is independent of the direction and speed of the motion vector. Since it can be uniformly obtained as a motion model such as (expressed by 3 field data), the force of the beef model and the deviation of 4 m can be calculated by 1 ton of flowers.
All you have to do is correct it.

Therefore, according to the present invention, the compression ratio can be increased.

First, since the correction is performed three-dimensionally, there are no problems with block distortion or anchored background.

Furthermore, since this invention uses pixel data of past fields with random noise superimposed as reference data when identifying parameters as prediction coefficients, it is possible to avoid the situation where an inverse matrix cannot be obtained. , parameters can be stably identified.

[Brief explanation of drawings]

FIG. 1 is a block diagram of an embodiment of the present invention, FIG. 2 is a block diagram showing a configuration for receiving encoded transmission data according to an embodiment of the present invention, and FIGS. is a route map used to explain one embodiment of the present invention. Explanation of main symbols in the drawings 1: Parameter identification section, 2: Prediction section, 3.4° 5: 5 field memories Two white noise generation circuits. Agent Patent Attorney Tadashi Sugiura Knowledgeable @11 Shoulder = bottom ゛aabcd

Claims (1)

[Claims] A memory that stores pixel data of several past fields, a linear linear combination of pixel data of the current field, and pixel data of the past several fields stored in the memory with random noise superimposed on it. and means for predicting pixel data of the current field from pixel data of the past several fields based on the identified parameters. 1. A high-efficiency encoding device for a television signal, characterized in that the above-mentioned identified parameters are transmitted.