JPS63197185A - Orthogonal transformation coding system - Google Patents

Abstract

PURPOSE:To realize high encoding efficiency in spite of the state of an image, by selecting an orthogonal transform data of minimum differential quantity by an arithmetic means, and coding the differential signal. CONSTITUTION:By taking a difference between the orthogonal transform data and plural data by differentiators 103-106, and selecting the data with the highest efficiency by a selector 108 based on prescribed reference, coding as a differential data is applied. In other words, plural data are the orthogonal transform data of the same block of all of the frames, the orthogonal transform data of the partial block of a present frame, the orthogonal transform data of a background image consisting of a temporally integrated background image or an image block at a specific time, and the orthogonal transform data at the specific time. Thus, by using the orthogonal transform data consisting of the temporally integrated background image or the image block at the specific time, the differential data goes to a very small value when a certain image part is returned to its original position in the reciprocal movement of the certain image part, then, it is possible to improve the coding efficiency more remarkably than the orthogonal transform data at the same position of all of the frames.

Description

[Detailed description of the invention] [Purpose of the invention] (Industrial application field) The present invention relates to an orthogonal transform encoding method.

(Prior Art) As a method for encoding TV moving images, there is an orthogonal transform encoding method in which a field or frame image is divided into blocks and orthogonal transform is performed.

The orthogonal transform differential encoding method, which encodes the difference between orthogonally transformed data and the orthogonally transformed data of the previous field (or previous frame), is an excellent method that can perform highly efficient encoding when the amount of motion is relatively small. It is a method.

The principle of this orthogonal transform differential encoding method will be explained using FIG. 2. The image signal divided into blocks of 8×8 samples is subjected to two-dimensional cosine transformation in a cosine transformation circuit 202. A difference between this output and the output of a frame memory 205 that holds the contents of the previous frame is calculated by a subtractor 203, and the output is converted into a matrix by a quantizer 204 and sent to an output terminal 208. On the other hand, the dequantized encoded data is sent to an inverse quantizer 20.
7 and is dequantized. As a result, the same signal as the input signal to the m-converter 204 is obtained, and this signal is added to the signals of all frames by the adder 206 and stored in the frame memory 205.
The contents of the frame memory 205 are rewritten.

In this method, when the amount of motion is small, the data after the difference is almost O, so it can be encoded with an extremely small number of codes m, but when the amount of motion is large, the number of codes increases rapidly. This has the problem that encoding efficiency does not improve.

In addition, in differential encoding methods that take differences without performing cosine transformation, it is possible to reduce the amount of information after the difference by performing hook compensation for the parallel movement of image components, but when it comes to two-dimensional cosine transformed data, On the other hand, there is also the problem that such motion compensation cannot be applied as is, and encoding efficiency cannot be improved.

(Problems to be Solved by the Invention) As described above, the conventional orthogonal transform encoding method has the drawback that the encoding efficiency is low depending on the image.

Therefore, it is an object of the present invention to provide an orthogonal transform encoding method that can achieve higher encoding efficiency regardless of the changing state of an image.

[Structure of the Invention] (Means for Solving the Problems) This invention calculates the difference between orthogonal transformed data and the following plural data, selects the most efficient data based on predetermined evaluation criteria, and calculates the difference. It is used as data and encoded. The plurality of data refers to orthogonal transformation data of the same partial block of all frames, orthogonal transformation data of another partial block of the current frame, orthogonal transformation data of a temporally integrated background image or a background image consisting of an image block at a specific time. conversion data,
This is orthogonal transformation data of specific data.

(Function) By using orthogonal transformation data of an image consisting of a temporally integrated cost image or an image block at a specific time,
When a certain image part makes one reciprocating motion and returns to its original position, the difference data becomes a significantly smaller value, and the encoding efficiency is significantly improved compared to orthogonal transformed data at the same position in all frames.

Also, by using orthogonal transformation of specific data, the start-up speed during large changes such as scene changes can be improved.

Also, the orthogonal transform differential encoding method has the disadvantage that it cannot perform motion compensation like the time domain differential encoding method, but on the other hand, since it is orthogonal transform data, it requires less memory to hold it compared to the time domain. It has the following characteristics.

For example, when the average value of all images, that is, the intermediate level value, is used as the specific pattern, all of the orthogonal transformation data except one DC component becomes 0, so no special memory is required.

Furthermore, even if data that is not all the same specific pattern data is used, the orthogonal transformation data is often concentrated in the low range, so the amount of data can be significantly reduced compared to retaining data in the time domain. It has its advantages.

These four types of orthogonal transformation data are the orthogonal transformation data at the same position in all frames, which is a still image or semi-still image, and the orthogonal transformation data at another position in the current frame, which is time-integrated in the case of a background or a specific time. Image orthogonal transformation data corresponds to reciprocating motion, and specific pattern orthogonal transformation data corresponds to scene changes and large movements. It has an effective effect on moving images with limited movement patterns, such as in TV conferences.

(Example) An embodiment of the present invention will be described in detail below with reference to the drawings.

FIG. 1 is a block diagram of an orthogonal transform four-way differential encoding system according to an embodiment of the present invention.

An input image signal inputted from an input terminal 101 and divided into blocks of 8×8 pixels is subjected to two-dimensional cosine transformation in a cosine transformation circuit 102 . This cosine transformed data is subtracted from four types of data by subtractors 103, 104, 105° and 106.

(2) A subtractor 103 performs a difference with orthogonal transformed data of 10 identical portions of all frames. (2) In the subtractor 104, the data is differentiated from orthogonally transformed data of another partial block of the current frame. (2) In the subtractor 105, the data is subtracted from the orthogonal transformed data of the data corresponding to the background.

(2) A subtractor 106 subtracts specific data, for example, orthogonal transformed data of one constant level of luminance.

The arithmetic unit 107 measures the magnitude of this difference data and selects the smallest difference data in a selection 11108. m
The weight of the difference data in the ti unit 107 may be determined by calculating the sum of the absolute values of 64 pieces of difference data, or by calculating the sum of squares.

The difference data selected by the selector 108 is sent to the quantizer 109
is quantized and output. In other words, after converting each 8x8 pixel block of the input image signal into the frequency domain (cosine transform), there are four prediction methods. A method of transmitting differences in information ■A method of storing information about the background in advance and transmitting only other information■
This method selects and quantizes the method that transmits the information most efficiently from among the methods of transmitting the difference from a certain level of luminance.

On the other hand, the output is dequantized by a quantizer 110. The inverse quantization output is added to the output of the cosine transform data memory of all frames in the adder 112 and then sent to the frame memory 11 again.
It will be written to 1. The cosine transform data of all the frames that move from moment to moment in the frame memory 111 are sent to the difference unit 110.
Sent to 3.

At the same time, background memory 115 is
Ill be reduced to integrated average M1 image data. This data is placed in an arithmetic circuit 116 and subjected to a nonlinear operation.

This performance circuit 116 has a comparison table of nonlinear characteristics as shown in FIG.
It changes linearly for the signal.

However, for signals input beyond a predetermined range (between A and 8), a trick is performed in which the change is made extremely small.

For example, when an image including a person moves slightly (between △ and B), a corrected image is input from the adder 117 to the background memory 115 by following changes in light shadows and the human face. Then, the contents of the memory 115 are rewritten.

However, if the person makes a large movement, the background memory 115 is not corrected and the previous data is retained. Performing such a calculation is effective for a movement in which a person returns to the original position, such as a swaying movement from side to side. In other words, this is because such a movement does not need to be regarded as a movement.

However, if the person does not return to their original position,
The contents of the background memory 115 are gradually rewritten.

As described above, the output of the arithmetic circuit 116 is added to the output of the fv scenery memory 115 by the adder 117 and written to the background memory 115 again. Background data in background memory 115 is sent to difference 105. Furthermore, m-child data from the inverse quantizer 110 is sent to the memory 113 and sent to the subtractor 104 as all blocks of the current frame. As the cosine transform data in the current frame, a block immediately before or directly above the block currently being processed can be used. The memory 113 can be implemented by a memory or a delay device such as a shift register.

It is sufficient to store blocks of 8×8 pixels in the specific pattern memory 118 as data obtained by two-dimensional cosine transformation of image data. As the specific pattern, a natural pattern as an image signal may be adopted. For example, if all pixels are set to the same average value level, all but one DC component will be O, so the memory 118 will become virtually unnecessary.

Note that the above embodiment is characterized by taking the difference using four types of orthogonal transformation data, but the four differentiators 103 to 1
06 may be shared. Further, although one pattern is used as the specific pattern, two or more patterns may be used. At this time, it is natural that the difference between the five orthogonal transformed data is taken, and the orthogonal transformed data used for the difference is not limited to four.

Further, the structure of the frame memory 111 to the background memory 115 may be made as shown in FIG.

In this case, the data in the frame memory 111 is directly input to the background memory, but the input timing providing means 501 gives a timing of, for example, once every 5 seconds to the background memory 111.
This involves rewriting the contents of 15. The input timing unnecessary means 501 divides a reference clock of, for example, 64 K bps by a predetermined frequency to produce a timing pulse of 15 seconds per pulse.

Next, another embodiment will be explained using FIG. 3. In this embodiment, after the difference as in the above embodiment is applied to the signal before cosine transformation, cosine transformation is selectively performed.

In this embodiment, the four types of data are data of the same partial block of all frames, data of another partial block of the current frame, data indicating the background, and specific data,
These are all data that have not been subjected to cosine transformation.

Background memory 316 stores data corresponding to the background. The memory 317 stores data of the current frame, and data of a block different from the block supplied to the input terminal 301 is supplied to the subtractor 304.

The frame memory 315 stores data one frame (or more) before the timing at which data is supplied to the input terminal 301 as a reference.

Here, data is extracted from the same block one frame before the data supplied to the input terminal 301 and supplied to the subtractor 303 . Specific data is stored in the memory 318 and is supplied to the subtractor 305.

The outputs of the four types of difference circuits 302 to 305 are sent to an arithmetic circuit 307.
After the selection circuit 306 makes a selection using the selection signal obtained in the above, a cosine transformation circuit 308 performs cosine transformation. That is, the magnitude of the four types of difference outputs is determined and the smallest difference value is sent to the cosine conversion circuit 308. In this case, when the level of the selected difference signal - for example, the sum of the absolute values - is sufficiently small, dumpling is performed directly without going through the OCT 308. A second selector 309 makes this selection. In other words, the selector 30
6, one of two different signals is selected by a selector 309 using a selection signal from the arithmetic circuit 307.

A quantizer 310 quantizes the selected signal. The quantized data sequence is sent to output terminal 319.

One-way quantized data is inversely quantized by an inverse quantizer 311, inversely cosine-transformed by an inverse cosine transform circuit 312, and selected by a third selector 313 to selectively update the contents of the frame memory 315. The frame memory output is sent to block memory 317 and background memory 316.

In this configuration, the level of the difference signal is generally smaller than in the case of a simple motion compensation difference or a background memory difference, so the number of orthogonal transformations of the difference signal is reduced, so it can be used even if the orthogonal transformation circuit has a slow performance speed. become.

In other words, by performing the four-way difference, two effects are produced: the difference signal level is reduced, the encoding efficiency is improved, and there is more leeway in the design of the orthogonal transform circuit.

In addition, the second selector 309 outputs two different difference signals (the signal directly input to the selector 309 and the signal input to the DCT3
08), the selector 306 not only judges the magnitude of the four types of difference outputs, but also judges whether the level of the selected difference signal is sufficiently small. You may have them do the same thing as 17JWk.

In this case, the second selector 309 is unnecessary. Furthermore, it is clear that the 8x8 pixel input image in all of the above embodiments may also be 16x16 pixels.

[Effects of the Invention] According to the present invention, excellent coding efficiency is exhibited even for reciprocating motion and scene engineering, for which the conventional cosine transform differential coding method did not improve efficiency. In addition, cosine transform encoding is a coding method that is performed in the time domain and has higher encoding efficiency than intra-frame interframe composite differential encoding, but it has the disadvantage that motion compensation prediction processing cannot be performed easily. According to the present invention, the coding efficiency is improved, so that the effects of the cosine transform coding method can be fully exhibited. Also,
In the present invention, since the frame memory is formed in the form of orthogonal transformation data such as cosine transformation, the memory capacity can generally be reduced compared to a frame memory in the time domain.

[Brief explanation of the drawing]

FIG. 1 is a diagram showing a cosine transform four-way differential encoding circuit according to an embodiment of the present invention, FIG. 2 is a diagram showing a conventional cosine transform differential encoding circuit, and FIG. 3 is a diagram showing another embodiment of the present invention. A diagram for explaining an example, the fourth factor is a diagram showing nonlinear characteristics of an arithmetic circuit, and FIG. 5 is a diagram showing a modification of FIG. 1. 102...Cosine conversion circuit 103.104, 105.106...Differentiator 1
08...Selector 111...Frame memory 113.118...Frame memory 115...
...Background memory

Claims (7)

[Claims] (1) A means for extracting a video signal in units of fields or frames, dividing it into blocks, and performing orthogonal transformation, and orthogonal transformation data of other blocks in the same field or frame and orthogonal transformation data of a certain block of a past field or frame. A difference means for performing a difference between the transformed data and four orthogonal transform data, that is, orthogonal transform data of a block having specific data and orthogonal transform data of a background image, a calculation means for calculating a difference amount from the difference data, and a calculation means for calculating a difference amount from the difference data; 1. An orthogonal transform encoding method, comprising means for selecting orthogonal transform data having a minimum difference amount by the means and encoding this difference signal. (2) The orthogonal transform encoding method according to claim 1, wherein the difference between the orthogonal transform signal of the image and the four orthogonal transform data is performed by excluding one of the DC components. (3) The orthogonal transform encoding method according to claim 1, wherein the specific data is all 0. (4) The orthogonal transform encoding method according to claim 1, wherein the orthogonal transform data of the background image is stored in a time-integrated background memory. (5) The orthogonal transform encoding method according to claim 2, wherein the orthogonal transform is performed after the difference. (6) The orthogonal transform encoding method according to claim 4, wherein the time integration is a nonlinear transform. (7) The orthogonal transform encoding method according to claim 4, wherein the background memory is an orthogonally transformed image block at a specific time.