JPH02119487A - Predictive coding method - Google Patents

Abstract

PURPOSE:To reduce prediction error by making the amplitude information of a frequency component in a block extracted for prediction constant and shifting the phase information by a change of position between the extracted block and the unextracted block. CONSTITUTION:The amplitude information of the frequency component in the extracted block for prediction is constant and the phase information is shifted by a change of position between the extracted block and the unextracted block. That is the correlation between picture elements utilizes the correlation between amplitude information and phase information while taking notice of a frequency component of a video signal. Since the predictive sensitivity in the prediction coding is improved, that is, the prediction error is decreased, then quantized noise is reduced, and the compression of a video signal is attained without causing production of gradient overload or edge business and picture quality deterioration such as deteriorated resolution.

Description

DETAILED DESCRIPTION OF THE INVENTION Field of Industrial Application The present invention relates to predictive coding of pixel information of a video signal.

2. Description of the Related Art There is a predictive coding method as one of conventional video signal compression methods.

This is a method in which the current pixel value is predicted from the past pixel value, the error between the predicted value and the actual value is determined, and this prediction error is used as a sign. At this time, by taking advantage of the statistical properties of the video signal that the correlation between adjacent pixels is high and the difference between them is small, nonlinear quantization such as logarithmic compression is performed, and the redundancy of the video signal is used for transmission. Alternatively, the amount of information to be recorded can be reduced. Methods for obtaining predicted values include previous value prediction using the immediately preceding pixel and two-dimensional prediction using pixels on the previous scanning line in addition to the immediately preceding pixel.

FIG. 2 is a sampling pattern diagram for explaining a two-dimensional predictive encoding method, which is a conventional predictive encoding method. In the figure, the O mark represents a pixel, A is the pixel and pixel value to be predictively encoded, and B, C, and D are pixels adjacent to the drawing A and their pixel values. Here, the predicted value A' for pixel A is calculated by calculation of the drawing values B, C, and D (predicted calculation or predicted numerical arithmetic), and the prediction error is A-A'.
Let be the sign.

Prediction calculations include, for example, a method of taking the average value of pixels B, C, and D, and can be expressed by the calculation A': (B+C+D)/3 (1'). Also, pixel A, pixel B, C,
There is also a prediction calculation based on A'=(3B+2C+3D)/8 (2'), which is weighted in consideration of the distance between pixels from D.

Comprehensive compression processing is performed by encoding the prediction error obtained in this way using the nonlinear quantization described above.

In the conventional predictive encoding method described above, the pixel arrangement of the video signal is in a matrix, but the same processing can be performed on a video signal with a non-matrix pixel arrangement as shown in FIG.

An example of a video signal with this non-matrix pixel arrangement is seen in the MUSE signal, which is scheduled to be used as a high-definition television signal transmitted to general households through satellite broadcasting.
The pixel array of one screen of the MUSE signal is in the form of a quincunx with an offset between lines. Also,
The subsample processing, which is also used for the MUSE signal, is also a method used as preprocessing when compressing high data rate video signals. In this respect as well, it is effective that the predictive encoding method can compress a video signal having a non-matrix pixel arrangement.

Problems to be Solved by the Invention However, the above method is effective when the correlation between adjacent pixels is high, but when the correlation is low, that is, when the prediction error is large, a large prediction error occurs due to nonlinear quantum As a result, quantization noise also increases, which causes problems such as gradient overload and edge business during decoding, and deterioration of image quality such as a decrease in resolution.

Furthermore, in the above-mentioned subsampled video signal, the correlation between pixels becomes lower due to aliasing due to the subsamples, so the above problem becomes a serious problem.

In view of the problems of the conventional technology, the present invention focuses on the frequency component of a video signal to determine the correlation between pixels, not just the value, and uses the correlation of the amplitude information to improve the prediction sensitivity in predictive coding. The object of the present invention is to provide a predictive coding method that has a good prediction error, that is, a prediction error that is small.

Means for Solving the Problems The present invention divides pixel information of one screen into a plurality of blocks each in a matrix shape, extracts an arbitrary block from the plurality of blocks, and calculates the frequency components in the extracted blocks. The pixel information of the unextracted block is predicted based on the pixel information of the unextracted block, and when making the prediction, the amplitude information of the frequency component in the extracted block is constant and the phase information is the same as that of the extracted block and the unextracted block. This predictive encoding method is characterized by shifting by the amount of change in position with respect to the block.

Effect: The present invention uses the above-described method to focus on the frequency component of the video signal, rather than just the value, of the correlation between pixels, and utilizes the correlation between the amplitude information and phase information. Since the prediction sensitivity is good, that is, the prediction error is small, the quantization noise is also small, and it is possible to compress the video signal without causing gradient overload, edge business, etc., or image quality deterioration such as a decrease in resolution during decoding.

Example Embodiments of the present invention will be described below based on the drawings.

FIG. 1 is a sampling pattern diagram for explaining one embodiment of the present invention. In Figure 1, mark ○ and
The marks represent pixels arranged within one screen of the video signal, and the pixel arrangement across the screen is non-matrix. Therefore, first, the pixels on the screen are divided into the O-print pixel group and the pixel group, each of which has a matrix-like pixel arrangement, and then each pixel group is divided into an n*n matrix-like pixel arrangement ( (n is a positive integer). 1st
In the figure, 1 is a -block in the circle-print pixel group, and the solid line represents the block range, and 2 is a -block in the -print pixel group, and the broken line represents the block range. Also, at this time, the order of the matrix in the block is n
is 4, and the number of pixels in the block is 16.

After dividing the video signal of the non-matrix pixel array into matrix blocks as described above, prediction between the blocks is performed. Here, it is assumed that block 2 is predicted using block 1. As a prediction method, it is assumed that the amplitude information of the frequency components included in block 1 is constant, and the phase information is that the phase of each frequency component is changed by the change in position from block 1 to block 2.

Next, a prediction function for realizing the above prediction method will be explained.

Discrete cosine transform is a method for frequency decomposing a blocked signal. Let C be an n zero 〇 discrete cosine transformation matrix, and the ('yj) elements c++, 4+ (where I, J"0.I+..., n-1) of C are C(:
, r + =dtcO5(J(i+1/2) n
/n) (1). Here, d is: The output matrix Y obtained by performing the two-dimensional discrete cosine transform on the input pixel block X is determined by the following equation.

Y = CTXC (3) (3) Each element cj i + 41 of equation Y indicates amplitude information of horizontal and vertical two-dimensional frequency components. However, since the phase information is identified in the equation (3) transformation, a transformation is performed to shift the phase for each frequency component.

The discrete cosine transform matrix is: CCT=CTC=E (unit matrix) (4), so the inverse discrete cosine transform of equation (3) is:
It is expressed as CYC■ (5). Equation (5) can also be expressed as “ ” 7X :l”−aV+:, 4iCE
14c"=7Xχ,'y,1.3+F+4
(6) However, the elements of E i are as follows. Using the transposed matrix c1 of C and C (k,
1=0,1. ..., row 1), and the elements of F are a column vector indicating a single frequency component whose phase advances by iπ/n per pixel in the vertical direction, and a column vector whose phase advances by iπ/n per pixel in the horizontal direction. is the product of a row vector representing a single frequency component that advances by jπ/n. Therefore, by giving an offset to the phase of each frequency component of these column vectors and row vectors, it is possible to represent a two-dimensional frequency component in which the frequency and S width are the same and only the phase is shifted.

Now, using the real number S as a variable, the (1°J) element of the nun matrix P5 is expressed as ps++, 4+ = dtcO5(!(j+1/2+S
)7[/n) (9) When the predicted pixel block Xab is moved from the input pixel block X by a pixel in the vertical direction and b pixels in the horizontal direction (a and b are arbitrary real numbers), the predicted pixel block Xab is , (6) , (7) ,
(s) From the formula, X ab =: ',:≦:
] γ2,' y(+ , , IP a d E+4Pb=
P-"YPb (10) Also, from equation (3), equation (10) YesX -
b ” P −'C' X CP b= (C
P−)”X(CPb) = R,”XRb (11). Here, nun with the real number S in equation (11) as a variable
The matrix R,s is a conversion matrix in which the amplitude information is constant and the phase information is shifted by S pixels.

From the above, the prediction function for obtaining the prediction block from block 1 to block 2 in FIG. 1 is a = b
= 1/2 (12) This is the above formula (11).

As explained above, in this example, the prediction method is
Instead of the conventional correlation between pixel values, the correlation between amplitude information of frequency components included in a video signal is used.

Therefore, it is possible to sufficiently predict high frequency components that could not be predicted based on pixel value correlation alone, increasing prediction sensitivity in predictive coding and reducing quantization noise in nonlinear quantization in the subsequent stage. can. It is possible to compress video signals without causing image quality deterioration such as gradient overload, edge business, or a drop in resolution during decoding.

As described in detail, according to the present invention, the correlation between pixels is focused not on mere values, but on the frequency components of video signals, and the correlation between amplitude information and phase information is utilized. To,
The prediction sensitivity in predictive coding is good, that is, the prediction error is small, so the quantization noise is also small, and video signal compression that does not cause image quality deterioration such as gradient overload, edge business, or resolution drop during decoding. This has great practical effects.

[Brief explanation of drawings]

FIG. 1 is a sampling pattern diagram for explaining a predictive encoding method according to an embodiment of the present invention, and FIGS. 2 and 3 are sampling pattern diagrams for explaining a conventional predictive encoding method. l... Block 2 in the O print pixel group... Block No. 2 in the print pixel group ○ ○ ○ ○ ○ ■ ■ ○ ○ ■ ■ ○ ○ ○ ○ ○ ○ ○ ○ ○ ○ ■ ■ ○ ○ ■ ■ ○ ○ ○ ○

Claims (1)

[Claims] Divide the pixel information of one screen into a plurality of blocks each in a matrix shape, extract any block from the plurality of blocks, and extract the pixel information of the blocks that were not extracted based on the frequency components in the extracted blocks. A predictive coding method for predicting, wherein amplitude information of frequency components in the extracted block is constant and phase information is only a change in position between the extracted block and the unextracted block. A predictive encoding method characterized by shifting.