RU2467499C2 - Method of compressing digital video stream in television communication channel - Google Patents

Abstract

FIELD: physics.

SUBSTANCE: invention relates to digital television and particularly to compression of a digital video stream in a television communication channel. A field sequence is divided into groups of three types: O-fields, which act as the reference; N-even fields, which are encoded via prediction based on the previous field within one frame; M-odd fields, which are coded with prediction based on the previous odd field from another frame. To improve visual quality of the image according to the disclosed method, odd and even fields change places in the entire video sequence or group of frames. Frames with a higher image definition are formed as a result. To increase efficiency of compressing a digital stream, the disclosed method employs cutting of readings in N-fields horizontally and vertically with subsequent restoration thereof at the receiving side, wherein two neighbouring N-fields form a full half-frame (field).

EFFECT: communication channel bandwidth reduction, as well as high efficiency of digital conversion of a video signal, lying in additional cutting of the volume of digital information and increasing visual quality of the image, with sufficiently easy implementation.

13 dwg, 1 tbl

Description

The present invention relates to the field of digital television and can be used to compress the digital video stream in a television communication channel.

The problem of reducing the digital stream in television remains relevant today. The wide frequency band occupied by the digital signal is the main obstacle when transmitting it through communication channels. At the same time, digital methods make it possible to create a new type of TV equipment, more reliable, stable, compact and technologically advanced.

Currently, more than 100 methods for converting an analog image signal to digital are known. All methods are based on the use of one of the types of modulation: pulse-code (PCM), differential pulse-code (DPCM) and delta modulation (DM). In addition, the methods differ in the type of processing of the analog signal before analog-to-digital conversion (ADC) and the type of subsequent processing of the primary digital signal. At all stages of the signal conversion, the properties of the image and the visual analyzer are taken into account, which allows to achieve maximum efficiency of the signal conversion process as a whole.

Reserves to reduce the digital stream without compromising the quality of the reproduced image are contained in the specifics of the TV signal, which has significant information redundancy, which is conditionally divided into statistical, physiological (psychophysical), spatial (intraframe) and temporal. It should also be noted that there are no sharp boundaries. Psychophysical and static redundancies are associated with temporal and spatial redundancies.

An example of a reduction in psychophysical redundancy can be a method of transmitting gradations of brightness from large and small parts (intraframe redundancy) and features of the perception of moving objects - temporary (interframe redundancy).

In most cases, video sequences (TV image) contain redundancy in two directions - temporal and spatial. The main statistical property on which the compression equipment is based is the inter-element correlation, which includes the assumption of the correlation of successive frames of video data. Thus, the values of individual elements of the image can be predicted either by the values of the nearest elements within the same frame (intraframe coding), or by the values of elements located in the nearest frames (interframe coding and motion compensation).

In some cases, for example, when changing the plot, the temporal correlation between the closest frames is very low. In such cases, the intraframe correlation, that is, the spatial correlation of image elements, plays a decisive role. However, if the correlation between successive frames of video data is high, it is desirable to eliminate interframe redundancy. Moreover, the applied image conversion methods are based on the fact that its digital equivalent (PCM signal) is reduced to a form convenient for reducing redundant information. The most effective method is to convert video information from the time domain to the spectral one. The conversion result is a combination of spectral coefficients that characterize the amplitudes of the spatial frequencies of the image.

Of the considered orthogonal transforms, the MPEG standard recommended the use of the discrete cosine transform (DCT), which is a special case of the two-dimensional Fourier transform.

DCT translates individual image blocks, the size of which is determined by 8 × 8 elements (or 16 × 16), from the spatial domain of the values of the temporary signal to the spatial frequency domain of the spectral plane of the coefficients (Fig. 1). As a result, an array of initial signal values (Fig. 1, a) corresponds to an array of the same number of coefficients representing the amplitudes of these cosine components (Fig. 1, b).

Two-dimensional DCT is carried out according to the formula:

Where

f _{(x, y)} - image samples with spatial coordinates x, y (from 0 to N-1);

N - image block size (N × N elements);

F _{(m, n)} - coefficients characterizing the image in the spectral plane m, n (from 0 to N-1).

DCT is reversible: according to the distribution F _{(m, n), the} inverse transform uniquely recovers f _{(x, y)} from the existing transformants in the decoding device:

Quantization is an important part of signal processing at which losses occur. It determines the accuracy of storing the results of DCT and the compression ratio. During quantization, each of the DCT values is divided by the quantization coefficient Q _{(m, n)} (Table 1), individual for each spatial frequency, which is taken from a predefined 8 × 8 quantization coefficient table (naturally, it should be the same for the encoder and decoder). This table can be taken by default or generated by the encoder for specific images and transmitted to the decoder along with the compressed data. For a color image signal, tables may vary for different components (Y, C _R , C _B ). When decoding, the original values are approximately restored by multiplying by the quantization factor. The foregoing is explained in Table 1.

Table 1 139 144 149 153 155 155 155 155 1260 -one -12 -6 2 -2 -2 2 144 151 153 156 159 156 156 156 -22 -17 -6 -3 -3 0 one -one 150 155 160 163 158 156 156 156 -eleven -10 -one 2 one -one -one -one 159 161 165 160 160 159 159 159 -7 -2 0 2 one 0 0 0 159 160 161 162 162 155 155 155 0 -one one one -one -one one 2 161 161 161 161 160 157 157 157 2 0 one -one -one 2 2 0 162 162 161 163 162 157 157 157 -one -one 0 -one 0 2 one -one 162 162 161 161 163 158 158 158 -3 one -3 -one 2 one -one -one Original Samples f _{(x, y)} DCT coefficients (rounded) f _{(m, n)} 8 eleven 10 16 24 40 51 61 158 0 -one 0 0 0 0 0 12 12 fourteen 19 26 58 60 55 -2 -one 0 0 0 0 0 0 fourteen 13 16 24 40 57 69 56 -one -one 0 0 0 0 0 0 fourteen 17 22 29th 51 87 80 62 -one 0 0 0 0 0 0 0 eighteen 22 37 56 68 109 103 77 0 0 0 0 0 0 0 0 24 35 55 64 81 104 113 92 0 0 0 0 0 0 0 0 49 64 78 87 103 121 120 101 0 0 0 0 0 0 0 0 72 92 95 98 112 one hundred 103 99 0 0 0 0 0 0 0 0 Quantum table Q _{(m, n)} The coefficient after quantization C _{q (m, n)}

1264 0 -10 0 0 0 0 0 142 144 147 150 152 153 154 154 -24 -12 0 0 0 0 0 0 149 150 153 155 156 157 156 156 -fourteen -13 0 0 0 0 0 0 157 158 159 161 161 160 159 158 -fourteen 0 0 0 0 0 0 0 162 162 163 163 162 160 158 157 0 0 0 0 0 0 0 0 162 162 162 162 161 158 156 155 0 0 0 0 0 0 0 0 160 161 161 161 160 158 156 154 0 0 0 0 0 0 0 0 160 160 161 162 161 160 158 157 0 0 0 0 0 0 0 0 160 161 163 164 164 163 161 160

Coefficients after inverse quantization

Recovered Sample Values

The table shows that the restored image differs slightly from the original.

The most famous image compression standards are MPEG-1, MPEG-2, MPEG-4, MPEG-7, JPEG, and wavelet transform. In television broadcasting, the MPEG-2 standard has found application. Other methods of image compression during experiments yielded worse results: poor quality, complexity of technical implementation, insufficient digital stream compression efficiency.

The elimination of intraframe and interframe redundancy in the MPEG standard is the most effective. It is almost impossible to change the standard in TV broadcasting, as this is associated with high economic costs. But in applied television and special digital television systems that are not related to accepted standards, the application of other methods of compressing the digital stream will not cause significant difficulties.

The MPEG-2 standard does not regulate video compression methods, but only determines what the bitstream of the encoded signal should look like, so specific algorithms are a trade secret of equipment manufacturers. However, there are general principles and the process of compressing a digital video signal stream can be divided into a number of sequential operations: converting an analog signal to digital form (ADC), preprocessing, discrete cosine transform (DCT), quantization and coding.

MPEG-2, unlike MPEG-1, provides the ability to process interlaced images. In MPEG-1, for encoding such images, it was necessary to first combine two fields into one frame and only after that send a signal to the encoder input. However, the procedure led to noticeable distortions such as blinds and a comb. MPEG-2 introduces the concept of field and frame coding. In field coding, two fields of one frame are encoded separately as separate images. Each field is divided into disjoint macroblocks (8 × 8) or (16 × 16) pixels (elements) and DCT is applied to them. Frame coding involves line-by-line combining of two fields into one frame and processing it as a regular image with progressive decomposition.

MPEG-2 defines two types of DCT for macroblocks: frame and field. Frame DCT acts similarly to MPEG-1: a macroblock of brightness samples of 16 × 16 pixels is divided into 4 blocks of 8 × 8 pixels in accordance with their location. Field DCT takes 8 lines from an odd field for the upper two blocks and 8 lines from an even field for the lower blocks, forming the odd and even fields of the macroblock, as shown in Fig.2.

Field DCT is more effective with a significant difference between the fields, for example, in the presence of vertical movement. Field coding can use only field prediction; with frame coding, the possibilities are wider — frame prediction or field prediction is allowed. In the latter case, the pairing of the separately odd and even fields of the macroblock in each of the two fields of the reference frame is searched for and the best result is selected.

Thus, in the MPEG-2 standard, the encoder and decoder interpret the video data as consisting of either 25 images per second with a constant vertical resolution, or 50 images per second with a half vertical resolution. Obviously, in both cases, the amount of information at the input of the encoder is the same. The field structure is better suited for compressing images with fast motion, providing fewer artifacts. That is, it is good for scenes with a lot of movement, but is less suitable for spatial redundancy, providing worse compression of still images in terms of minimizing artifacts. For a personnel or progressive structure, the opposite conclusions are true, respectively. The choice between personnel and field structures is carried out by a specialist to achieve the maximum level of image quality.

After pre-processing, difference prediction errors or macroblocks themselves are subjected to DCT, as a result of which the original matrix of blocks is converted into a matrix of coefficients (Fig. 3). The coefficients in the matrix are arranged in increasing frequencies in the vertical and horizontal directions. In other words, the DCT matrix can be considered as a two-dimensional filter in two directions. In this case, the main energy is concentrated near zero frequencies. After quantization of these coefficients, they are converted into a one-dimensional sequence. The algorithm for ordering the coefficients is that as a result of scanning the matrices are combined in a series with a description of their length and location (Fig. 3). One of the options for this algorithm is a zigzag scan, in which the conversion starts from the upper left corner and ends in the lower right. The quantization obtained after the DCT matrix is made taking into account the sensitivity of the eye to various spatial frequencies. Human vision is most sensitive to gradations of brightness of large parts, therefore, for coefficients corresponding to zero frequencies, the quantization step should be minimal. For higher spatial frequencies, quantization to a smaller number of levels is applied.

When constructing a digital stream compression system in a communication channel, it is necessary to take into account the peculiarities of vision when perceiving small parts of different lengths and orientations in the xy plane (Fig. 1). It is known that the transmission of thin vertical black and white lines through a communication channel requires a wide band of frequencies, and thin horizontal lines are transmitted at low frequencies, although the resolution of the eye when perceiving vertical and horizontal lines is the same.

This should be taken into account when reducing the digital stream, so that in some fields thin horizontal lines prevail, which are transmitted by low frequencies. These features allow you to further reduce the redundancy of the digital signal by thinning out the samples of the digital stream in the encoder. In the decoding device, the inverse operation is performed - interpolation, which restores the original number of samples.

Similar methods of reducing the digital stream are often used in television, for example, in wavelet transform.

MPEG-2 also uses inter-frame coding, which greatly improves the compression efficiency of the digital stream.

With inter-frame coding based on temporal redundancy, various prediction methods are possible. Depending on this image (frames) in their time sequence are divided into the following types:

- I-frames (intra), reference, are basic and are encoded without reference to other frames, that is, using information from this frame only. Type of coding - intraframe, providing moderate compression. Prediction for them is not formed. All other frames are analyzed by the processor, which compares them with the reference, as well as with each other.

- P-frames (predictive - predicted), the transmission of which uses inter-frame coding by prediction with motion compensation for the nearest previous 1-frame or P-frame. If the macroblock in the P-frame cannot be described using motion compensation, which happens when an unknown object appears, then it is encoded in the same way as the macroblock in the I-frame. P-frames are compressed 3 times stronger than I-frames, and serve as reference for incoming P and B-frames.

- B-frames (bidirectional - bidirectional), which are transmitted with inter-frame coding by prediction with motion compensation for I-frames and P-frames closest to them both in front and behind, and cannot be used to predict other frames (some fragments B-frames can be encoded using the intraframe method).

Thus, the MPEG-2 standard uses 3 types of coding: intraframe, interframe forward with motion compensation, bidirectional interframe also with motion compensation.

Received frames are combined into sequential frame groups (GOPs). Each sequence begins with an I-frame and consists of a variable number of P- and B-frames.

At the beginning of the plot there should be an I-frame, at the end of a B-frame. It is possible to increase the share of B-frames only within the framework of one plot, otherwise large errors of prediction and motion compensation will occur. Since the typical duration of a group of frames (in the temporary representation of 0.5 s - 12 frames) is significantly less than the characteristic distance between the boundaries of the plots, in most cases, rigidly setting the structure of the group of frames does not lead to significant visual errors due to the fact that the change of plot falls inside groups of frames.

Consider an example frame sequence:

one 2 3 four 5 6 7 8 9 10 eleven 12 13 fourteen fifteen 16 17 eighteen 19 twenty I AT AT R AT AT R AT AT R AT AT R AT AT I AT AT R AT

Here frames 1 to 15 form a group of frames. The number of frames in a group may be different, but they always start with an I-frame. P-frame 4 is predicted by I-frame 1, P-frame 7 - by P-frame 4, P-frame 10 - by P-frame 7, etc. The I-frame is transmitted with intraframe coding independently of all the frames preceding it. B-frames 2 and 3 are predicted from I-frame 1 and P-frame 4, B-frames 5 and 6 are predicted from P-frames 4 and 7, etc. B-frames 14 and 15 are predicted from I-frame 16 and P-frame 13. Before encoding, the order of the frames changes, since each B-frame must go after both frames over which it is predicted:

one four 2 3 7 5 6 10 8 9 13 eleven 12 16 fourteen fifteen 19 17 eighteen 22 I R AT AT R AT AT R AT AT R AT AT I AT AT R AT AT R

In this order, frames are encoded and transmitted, and in the decoding process, the original frame order is restored. The change in the transmission order of frames B, B, P is explained by the fact that when they are decoded at the receiving end of the system, frames I and P will be required.

Type B image is most deeply compressed. If P-frames require 3 times fewer bits for their transmission than I-frames, then in the images the number of bits for most scenes is 2-5 times less than in P. As a result, the noise immunity of B-frames suffers. Therefore, to protect against possible errors, images B are not used to predict other frames.

Of the considered methods of reducing (decreasing) the digital stream in the communication channel, which are the closest and the proposed technical solution, noteworthy is the method of transmitting and receiving digital information according to the MPEG-2 standard, which we take as a prototype of the invention. The above analysis of the MPEG-2 system shows that it also has disadvantages. It is known that the compression rate of a digital stream directly depends on the correlation between image elements. In MPEG-2, inter-frame coding between I- and P-frames is performed through 3 frames (fields), and not through 1 frame (field). This greatly weakens the correlation between the elements and, accordingly, worsens the effectiveness of reducing image redundancy. Although this drawback is compensated by the bi-directional prediction of B-frames, the noise immunity and image quality of B-frames are reduced.

Naturally, the permutation of P-, B1- and B2-frames significantly complicates the construction of the system, making it cumbersome and expensive. Moreover, the mentioned complications of the system do not always justify themselves in the sense of obtaining high image quality.

It should be noted that bidirectional prediction allows you to interrupt the error track that occurred in a group of frames if an error occurred in B-frames. In I- and P-frames, the error track will continue until the beginning of the next reference frame I.

An increase in the compression efficiency of B-frames is accompanied by a loss of image quality in these frames, but due to the higher image quality of I- and P-frames, our eye does not visually notice a deterioration in the quality of the entire image. In addition, there are certain difficulties in determining the motion vector of an object in adjacent frames due to a change in the sequence of B- and P-frames. At the same time, any changes to the MPEG-2 standard may lead to a decrease in coding efficiency.

Elimination of the aforementioned shortcomings will make it possible to create a simpler and no less effective digital stream compression system for applied and broadcast television.

As noted above, the closest technical solution (the prototype of the present invention) is a method of compressing a digital video stream according to the MPEG-2 standard, the work of which was discussed in detail above in the work (Lokshin V.A. Digital broadcasting: from studio to viewer. / Edited by L.S. Vilenchik. - M.: Cyrus system, 2001. - 448 p.)

The implementation of the prototype involves the following sequence of actions (relative to the brightness signal):

- the image is completed to a multiple of “16” the number of elements (pixels) in rows and columns to ensure that the image is divided into an integer number of macroblocks;

- the image is divided into a sequence of macroblocks, each of which consists of 4 blocks of 8 * 8 elements that carry information about the brightness of the object (figure 2);

- the formation of reference (I-frames (fields)) and differential (P-, B-frames (fields)) signals;

- determination of the motion vector of blocks;

- discrete cosine transform (DCT);

- quantization and coding of DCT coefficients;

- multiplexing digital stream data;

- regulation of the flow rate of data to be transmitted.

The previously noted disadvantages of the prototype impede its implementation in applied television.

The aim of the invention is to narrow the frequency band of the communication channel, as well as to increase the efficiency of digital video signal conversion, which consists in further reducing the amount of digital information with a fairly simple technical implementation.

The goal is achieved in that, as in the prototype, the sequence of fields (rather than frames) is divided into groups. In the group there are fields of 3 types (figure 4):

- O-fields - images playing the role of reference when restoring other images. Prediction for them is not formed, use in-field coding;

- N-even fields - images encoded by prediction based on the previous field within the same frame use inter-field coding, resulting in an inter-line difference between two adjacent lines of odd and even fields;

- M-odd fields - encoded with prediction based on the previous odd field from another frame.

The letters in Fig. 4 indicate their type, and the numbers 1, 2, 3, ... indicate the order of their arrival at the compressor input. There are 10 fields in the group (5 frames). The group begins with an image of type O, is subjected only to field compression. In images of N- and M-type fields, both spatial and temporal redundancies are eliminated. The number of fields (frames) in the group may be different than that shown in figure 4, and can be easily varied by changing the pulse repetition rate U ₁ and, accordingly, U ₄ . The arrows in FIG. 4 determine the prediction order: for N, within one frame, and M, within 2 adjacent frames for odd fields.

Images of N-fields mainly contain thin horizontal lines of different lengths (low-frequency components). When moving in images of N-fields, high-frequency components appear.

Images of M-fields contain mainly high-frequency components, reflect the inter-frame difference within the odd fields of adjacent frames. Here, the low-frequency components appear during a sharp change of plot.

When transferring inactive objects in images of M-fields, the signal is practically zero, and the signal from N-fields is non-zero. The RF components in N-fields appear from the transmission of oblique lines. After the formation of difference signals of N-fields in front of the DCT matrix, to further reduce the digital stream, we will perform decimation (decimation) of signals from N-fields in both horizontal and vertical directions. This operation is legitimate and does not impair image quality, since the sampling frequency for N-fields is too high. In the decoding device, we carry out the inverse operation - interpolation, restoring the lost information. Such operations are the basis for reducing the digital stream in wavelet transforms in the LF and HF channels.

The operations of thinning samples and interpolation can also be performed for signals from M-fields. In the decoding device, the restoration of the missed samples is carried out by using a one-dimensional digital filter with a finite impulse response.

Such devices are interpolators. Of course, the implementation of interpolation is associated with certain inaccuracies in the restoration of the signal from its samples, however, it is necessary for operations to obtain the initial values of the signals. If there is no interpolation in the decoding device, the high-frequency components of the difference signals may be lost, and interfering low-frequency components will appear in the video signal.

As a result of the transformations, the digital stream in N-fields additionally decreases by about 4 times.

As in the MPEG-2 standard, entropy coding and motion vector determination in M-fields are applicable in the compression system. In N-fields, motion compensation can be ignored, since they are located in the same frames as the O- and M-fields, in addition, it will disrupt the operation of the interpolator in the decoding device or significantly complicate its construction.

Summing up the above, we see that the efficiency of reducing the digital stream of O-fields is the lowest in comparison with the M- and N-fields and is approximately equal in efficiency to 1-fields in the MPEG-2 standard. But the image quality is higher than in the M and N fields. Compression efficiency in M-fields is approximately the same with P-fields, and in N-fields it is slightly higher than in B-fields, since in N-fields thinning (oversampling) of the values of the DCT matrix signals (to the coefficient matrix) in the horizontal and vertical directions (FIGS. 5 and 6).

In addition, to improve the visual quality of the image according to the proposed method, the odd and even fields in each group of frames are interchanged: in the 1st group - (O, N), (M, N), (M, N), ..., and in 2- group - (N, O), (N, M), (N, M), ... etc. (figure 4, a '). Such a permutation of the fields due to the peculiarities of visual perception of vision leads to a significant improvement in the visual quality of the image as a whole, that is, visually the formation of a reference frame (O, N) + (N, O) occurs. In the receiver, of course, the initial sequence of the fields should be restored.

Another variant of the permutation of fields in frames (Fig. 4, a``) is the permutation of odd and even fields in the entire video sequence of frames: (O, N), (N, M), (M, N), (N, M ), (M, N), (N, O), ... etc. With an odd number (2n + 1) of fields in the group of frames, the O- and N-fields, as well as (M- and N-fields), are interchanged periodically, which is more preferable. Here a virtual reference frame is visually formed: (O, N) plus (N, O). In our case (Fig. 4, a), visually the frames (O, N) and (N, O) overlap each other through 4 frames, forming a reference frame.

Of the options considered, the latter is preferable, since the frequency of the field permutation is higher than in the first option.

That is, if the viewer is presented alternately with frames (fields) of images with high and low definition, then the eye as a whole perceives the image with high definition. In the prototype, this possibility of increasing the visual quality of the image is missing.

The implementation of the method involves the following sequence of actions (relative to the brightness signal):

- the image is extended to a multiple of “16” the number of elements (pixels) in rows and columns to ensure that the image is divided into an integer number of macroblocks:

- 1st option: rearrangement of odd and even fields in each group of frames to improve visual image quality: in 1 group - (O, N), (M, N), (M, N), ..., and in 2- group - (N, O), (N, M), (N, M), ... etc. (figure 4, a ');

- 2nd option: rearrangement of odd and even fields in the entire video sequence of frames: (O, N), (N, M), (M, N), (N, M) ... (N, O), (M, N), (N, M), (M, N), ... etc. (Fig. 4, a``);

- the image is divided into a sequence of macroblocks, each of which consists of four blocks of 8 * 8 elements that carry information about the brightness of the object (figure 5) for frame and field coding;

- the formation of reference (O-fields) and difference (M- and N-fields) signals;

- thinning of samples in the matrices of values of the differential brightness signals in rows and columns in the even fields of each frame (Fig.6 and 8);

- discrete cosine transformation of the values of the signal matrix into a matrix of coefficients;

- quantization and entropy coding of DCT coefficients;

- determination of motion vectors of blocks in a moving image and entropy coding;

- multiplexing digital stream data;

- regulation of the flow rate of data to be transmitted;

- decoding of received messages occurs in reverse order to the processes occurring in the encoding device.

Comparison of the proposed solutions with the prototype revealed signs that distinguish the claimed solution from the prototype, which allows us to conclude that the proposed method meets the criterion of "significant differences".

Significant differences of the proposed method are:

- apply for each group of frames simultaneously field and frame coding. MPEG-2 uses only frame or only field coding, depending on the nature of the transmitted image, which greatly complicates the equipment;

- a much stronger correlation between the compared signals, and therefore, a more effective elimination of redundancy from television messages. In the prototype, the correlation between the compared signals is much weaker than in the proposed method;

- thinning of samples in the matrices of values of one of the difference brightness signals along rows and columns in even fields of each frame;

- rearrangement of odd and even fields in the entire video sequence of frames to improve visual image quality.

Various circuit solutions are possible that provide such a reduction in the digital stream in the communication channel.

Consider the implementation of the method on the example of the device depicted in Fig.7 (encoding device) and Fig.9 (decoding device).

Briefly consider the operation of the encoder.

The block diagram of the encoder (Fig. 7) displays only the basic operations performed during encoding and providing the output data stream with the required parameters.

The encoder implements two encoding modes: intra-frame and inter-frame with prediction and motion compensation.

Note

To facilitate understanding of the physical processes taking place in the circuit, we will consider the video sequence of frames without rearranging the fields (Fig. 4, a).

The diagram shows only those nodes and blocks that characterize the features of this method of reducing the digital flow of information.

The operation of the circuit is also illustrated by the voltage diagrams presented in figure 4.

The input video signal from the PCM is fed to the signal preprocessing unit (BPO).

The following transformations are carried out in BPO signals:

- the image is completed to a multiple of “16” the number of elements (pixels) in rows and columns;

- rearrangement of odd and even fields in the video sequence of frames;

- the image is divided into a sequence of macroblocks, each of which consists of four blocks of 8 × 8 elements that carry information about the brightness.

Similar conversions are carried out with color signals in the 4: 2: 0 format. Next, in blocks 2, 3, 4, 5, 8, the formation of reference and difference signals M and N. The control signals U1, U2, U5 (Fig. 4) coordinate the operation of these blocks. In block 9, oversampling (decimation) of N-field signals in the horizontal and vertical directions is carried out.

All macroblocks of O-fields are encoded in intra-field coding mode. For this purpose, the PCM input signal is supplied to the DCT unit via the upper channel. In DCT 6 and 10, the matrices of the signal values f _{x, y} are transformed into the matrix of Fourier coefficients F _{(m, n)} . In the quantizer, the DCT coefficients are encoded in accordance with the formula:

where Q (m, n) are the quantization coefficients specified in the form of a table of 8 × 8 integers (table 1, Q);

ρ is a parameter that determines the degree of image compression;

Round - operation of rounding to the nearest integer value;

C _q (m, n) —the quantized DCT coefficients obtained as a result of this operation, which can be both positive and negative (Table 1, C _q ).

The quantization of the matrix of coefficients obtained after DCT is made taking into account the sensitivity of the eye to various spatial frequencies, and when transmitting more RF components, a large error is possible, that is, they can be quantized to a smaller number of levels. For large parts, DCT coefficients are quantized to a greater number of levels.

As a result of the operations of division and rounding, many DCT coefficients become equal to zero. It is quantization that makes it possible to reduce the number of binary symbols needed to represent information about the DCT coefficients, that is, image compression. At the same time, quantization is a source of irreversible loss of information during compression. The selection of a particular Q quantization table is at the discretion of the user.

In the blocks "Entropy coding" (21-23), coding is performed with a variable length of code words.

In the feedback loop (KV ^-1 dequantizer (15), inverse discrete cosine transform block - DCT ^-1 (14) and predictors (12 and 19), whose memory can contain several previous fields (τ = T _p + 1 / 2T _s , τ = 2 T _p )), the formation of the predicted field occurs.

The motion of OD (18) is estimated by comparing the current image fed to the input of the encoder with the image located in the memory (19) and used for prediction. Through block 16, the input of the OD (18) receives the signals of O- and M-fields.

The predictor in the system is not just a memory for storing previous fields (frames), but also a device that, when generating a prediction, searches from the array of data in its memory for a block that is consistent with the blocks of the current frame. For this, data on motion vectors are entered into the predictor. The prediction signal is also subjected to entropy encoding (23) and is multiplexed (24) into a common digital stream with DCT coefficients. The main digital streams O, M, N from the outputs of blocks 21 and 22 are multiplexed (24) into one stream. Signals from the output of the OD motion vector determinant are multiplexed into the common digital stream (18). In blocks 21, 22 and 23, an additional reduction of the digital stream is carried out.

The total digital stream after the multiplexer Mp (24) is fed to the input of the buffer memory BZU (26), which works according to the principle: "first in, first out." The need for introducing into the system BZU due to the following circumstances. Depending on the detail of the movement and the nature of the transmitted movement, the speed of the digital stream at the output of the variable-length coding units 21, 22, 23 can substantially change. With increasing level of RF components in the image, with rapidly changing scenes, the data flow rate at the compressor output increases. This increase can lead to exceeding the capacity of the transmission channel in its bandwidth. The rate limitation of the encoded digital stream is implemented by the implementation of feedback, which includes a buffer memory 26 and quantizers 7, 11.

To optimize the operation of the system, it is desirable to maintain the filling level of the CCD approximately constant. If the BZU is full, then data loss will occur, that is, image quality deterioration. If the BZU is freed, then "empty" blocks are transmitted through the communication channel, which leads to a decrease in the efficiency of the communication channel. Feedback the degree of filling of the CCD keeps constant.

The essence of feedback (OS) is as follows. If a fine-grained image is transmitted and the filling of the CCD increases (memory is full), then under the influence of the OS, the quantization parameter ρ of the DCT coefficients increases (formula 3). In this case, the number of bits per coefficient decreases, and the level of data flow is maintained approximately constant. On the contrary, when transmitting “smooth” images, quantization becomes more accurate (the parameter ρ in formula 3 decreases, and the quantization coefficients C _q - Table 1 - increase). This method corresponds to the properties of human vision: in fine-grained images, inaccuracies within brightness levels are less noticeable. Of course, a change in the quantization scale depending on the content of the image affects the quality of the reproduced image, the level of quantization noise changes. The UKS 25 compression ratio control device included in the feedback circuit increases its speed. As a result, due to the feedback action, the degree of filling of the buffer memory (BZU) is kept constant on average.

The most preferred circuits of the encoder of the ONM system is the circuit shown in Fig.7, where the thinning of the samples is carried out before DCT. The higher compression efficiency here is due to the stronger coupling between the elements in the block than between the DCT coefficients. The scheme of Fig. 7 provides higher image quality than the scheme where decimation is carried out after DCT. The choice of a particular coding scheme is determined by the technical data for the design of the system.

Thinning reports in comparison with the well-known has its own characteristics, which improve the visual clarity of the image of N-fields (Fig):

- in 2, 6, 10, ..., (4n-2) N-fields, only 2, 6, 10, ..., (4z-2) lines with half the number of samples (2, 4, 6, 8, ..., - thinning) (Fig. 8, c). Information in 4, 8, 12, ... lines is not transmitted;

- in 4, 8, 12, ..., (4n) N-fields, lines with numbers are transmitted (Fig. 8, d) 4, 8, 12, ..., (4z), and information in 2, 6, 10, ... lines not getting through. (Here n and z-serial numbers: 1, 2, 3, 4, 5, 6, ...).

This order of thinning samples in N-fields increases the visual clarity of N-fields to the original, since a virtual full-fledged N-half-frame (field) is created from two adjacent fields when using the 1st option of field permutation. It should be noted that as a result of thinning N-fields, the orthogonal sample structure is preserved, which is very important for DCT.

The decoding device (Fig. 9) performs the inverse process of converting a video signal from a digital form to an analog one. In the decoding device (Fig. 9), the operations of decoding codes of variable lengths (decoding the entropy code) 3, 6, dequantization 4, 7, inverse DCT (DCT ^-1 ) 5, 8 are performed, and the original sequence of fields with pulse-code modulation (PCM) is restored )

и τ=2Т_П (Т_П, Т_с - длительность поля и строки), аналогичные соответствующим блокам кодера.The decoder contains a buffer memory (CCD) 1, a demultiplexer (DMP) 2 for separating digital streams, decoders for codes with a variable length of code words (decoding of the entropy code) of the signals of the O, N, M fields, dequantizers (Qu ^-1 ) 4, 7, blocks of the inverse discrete cosine transform (DCT ^-1 ) 5, 8 and restoration of samples (↑ 2) 9, predictors and storage devices 12, 14 with a delay of signals on

and τ = 2Т _П (Т _П , Т _с - field and line duration), similar to the corresponding encoder blocks.

The clock frequency is restored using data from the decoding stream.

BZU at the input of the decoder performs the function of matching a constant bit rate in the input data stream with the processes in the decoder, that is, the original speed of the digital encoder stream is restored without taking into account the UKS adjustment (Fig. 7).

From the output of the buffer memory, the encoded image data and the values of the quantization parameters are sent to the decoding units of the entropy code 3.6 and then to the dequantizers (Qu ^-1 ) 4, 7, and the signals of the motion vectors are sent to the decoding unit of the entropy code 10 and the predictor 12.

As a result of the transformations performed at the output of the electronic switch EC 17, we obtain the initial sequence of O-, N- and M-fields with pulse-code modulation (PCM).

Let us consider in more detail the process of the formation of the initial signals of O-, M- and N-fields.

The digital signal from the output of the DCT ^{-1 -1} block contains the O-field reference signal transmitted with intraframe coding, and the M-field difference signal (interframe difference of two odd fields of adjacent frames). In the second channel, in addition to decoding the entropy code, dequantization, and inverse DCT 8, in the signal N, the missing samples are restored (interpolated) in block 9. The PCM is restored in the N and M signals as follows.

- одного поля и полстроки

) и через схему И 13 на предсказатель 12 (ЗУ с τ_З=2Т_П). При появлении сигнала N в сумматоре 15 происходит сложение его с задержанным опорным O-сигналом и образование ИКМ сигнала N, который через блок 16 поступает через ЭК 17 на выход схемы (переключатель в положение 2).When a signal O appears at the output of DCT ^-1 5, the electronic switch 17 passes the signal “O” to the output of the circuit (position 1 of the switch). At the same time, the signal O through the adder 11 is fed to the predictor 14 (memory device for a while

- one field and half a line

) and through the circuit And 13 to the predictor 12 (memory with τ _W = 2T _P ). When the signal N appears in the adder 15, it is added to the delayed reference O-signal and the PCM signal N is formed, which, through block 16, enters the circuit output through EC 17 (switch to position 2).

The signal M supplied to the adder 11 where it is delayed by the addition with τ _W = 2T _n reference signal "O". From the output of the adder 11, the restored M-signal from the PCM through the block 16 is fed through the EC 17 (switch to position 2) to the output of the circuit.

Thus, all three signals (O, N and M) with PCM, having passed the field rearrangement in block 18, acquire the initial sequence of following the fields O, N, M (Fig. 4, a) and can be applied to the DAC (in the diagram not shown).

Summing up the above, we can draw the following conclusions:

1. The proposed method compares favorably with existing methods for compressing a digital video signal stream in a communication channel. Not inferior in the efficiency of digital stream compression to the MPEG-2 standard, the proposed method is much simpler and easier to implement in practice. Known methods for compressing the digital stream MPEG-4, MPEG-7, wavelet transform are more complicated than the proposed method in the sense of technical implementation, therefore, they are limited in their application. In the prototype it is impossible to rearrange the fields to increase visual clarity: in the proposed method, the fields with high and low definition are swapped, this creates a virtual full frame. This is not the case in MPEG-2, since the permutation of frames (fields) is carried out only to obtain interframe (field) differences, while high and low definition frames (fields) always take their specific place in the group of frames during playback.

The image quality in the proposed method is higher than in the prototype due to the permutation of adjacent fields in the video sequence of frames.

2. If necessary, the digital stream compression system according to the proposed method easily switches to the “reference frame” mode of operation when a sequence of fields consisting of frames is transmitted: (O, N), (N, O), (O, N), ..., that is, the fields O and N in each frame are swapped. As a result of the permutation of the fields, the visual clarity of the image increases. All frames are visually reproduced with the image quality of the reference frame (virtual).

The transition of the system to the operating mode of the "reference field (frame)" in applied television can be carried out automatically by an alarm signal.

3. In the system built according to the proposed method, there are large reserves to improve the compression efficiency of the digital stream.

Consider an example of increasing the compression efficiency of a digital stream by the proposed method.

Let the PCM signal with the orthogonal arrangement of the samples in the frame be received at the input of the encoding device (Fig. 10, a). In odd M-fields, in order to maintain visual clarity, the sample structure is thinned once in the horizontal direction in a checkerboard pattern (Fig. 10, d): in 1, 5, 9, 13, ..., (4n-3) lines, only odd elements are transmitted, and in 3, 7, 11, 15, ..., (4n-1) lines - even elements. Such oversampling improves visual horizontal clarity. In even N-fields, the horizontal signal is thinned twice, that is, only every 4th element is transmitted along the line, but in a checkerboard pattern. All lines are transmitted vertically. The method of thinning and restoration of samples remains the same (Fig. 10, c, d) and the visual clarity of the image as a whole decreases slightly.

In our case, it is supposed to restore the sampling frequency of N-fields to the original value, so there will be no noticeable loss in image quality. Naturally, before thinning to reduce distortion, there is a low-pass filter, which is not shown in the circuit (Fig. 7) for simplicity.

The additional thinning of the samples in the considered case increases the efficiency of the system in reducing the digital stream by more than 2 times. In the prototype, this is not possible.

4. Significantly stronger than in the prototype, the correlation between the elements being compared ensures the effective elimination of redundancy from television messages and makes the system promising for its modernization.

Let us cite the schemes of the encoding and decoding devices that implement the proposed method of compressing the digital video signal stream to the form required when processing the application (Figs. 11 and 12).

The drawings indicate the following notation:

For figure 11:

- pre-processing unit - 1;

- subtraction device - 2;

- the key scheme of I-3, 4, 8, 16;

- addition device - 5, 13, 17;

- discrete cosine transform - 6, 10;

- quantizers - 7, 11;

- block thinning signals N - 9;

- predictors with τ ₃ = 1T _P and τ ₃ = 2T _P - 12, 19;

- block reverse DCT ^-1 - 14;

- dequantizer sq - 15;

- motion determinant - 18;

- multiplexer - 24;

- device for controlling the compression ratio - 25;

- buffer storage device - 26.

For Fig:

- buffer storage device - 1;

- demultiplexer - 2;

- decoder of the entropy code - 3, 6, 10;

- dequantizer - 4, 7;

- block inverse discrete cosine transform - 5, 8;

- block interpolation of samples in the channel of N-fields - 9;

- adders - 11, 15, 16;

- predictors: τ ₃ = 2T _P - 12, τ ₃ = 1T _P - 14;

- electronic switch - 17;

- block reverse permutation of fields - 18.

The diagrams show only those nodes and blocks that characterize the features of this method of compressing a digital video signal stream in a communication channel.

Consider briefly the sequence of transformations of the digital video signal in the presented schemes (Figs. 11 and 12).

The digital video signal from the PCM enters the signal preprocessing unit 1, in which the image blocks are formed and the odd and even fields in the sequence of frames are rearranged. From the output of block 1, the signal along the upper channel through the key circuit 3 and adder 5 is fed to the discrete cosine transform block of DCT 6. In subtracter 2, two difference signals of M- and N-fields are generated. The signal of the M-fields through the key 4 and the adder 5 is fed to the DCT block 6. In the DCT blocks, the matrix of signal values is converted to the matrix of coefficients of the Fourier series. Another N-field signal through key 8 and decimation unit 9 is fed to DCT 10. From DCT 6 and 10, signals O, M and N, having passed quantization 7 and 11 and entropy coding 21 and 22, are sent to multiplexer 24.

и 19 с τ₃=2Т_П, полученный задержанный сигнал обратной связи в блоке 2 вместе с сигналом, поступающем с БПО, образует разностные сигналы М и N. Кроме того, с выхода БПО 1 входной сигнал через ключевую схему 16 поступает на определитель движения 18, на второй вход которого поступает задержанный сигнал полей О и М и определяется вектор движения, который, пройдя блок энтропийного кодирования 23, поступает на вход мультиплексора 24. Далее объединенный в мультиплексоре цифровой поток поступает на вход буферного запоминающего устройства БЗУ 26. С помощью БЗУ 26, блока управления коэффициентом сжатия УКС 25, квантователей 7 и 11, мультиплексора 24 осуществляется выравнивание скорости цифрового потока в канале связи независимо от содержания передаваемого изображения. Этим самым повышается эффективность использования канала связи, пропускная способность которого ограничена. Скомпрессированный цифровой поток с выхода БЗУ 26 поступает в канал связи.In the feedback loop covering the blocks: dequantizer kv ^-1 15, block of the inverse cosine transform DCT ^-1 14, adders 13, 17 and 20, predictors 12 s

and 19 with τ ₃ = 2T _P , the received delayed feedback signal in block 2, together with the signal from the BPO, forms the difference signals M and N. In addition, from the output of the BPO 1, the input signal through the key circuit 16 is fed to the motion detector 18 , the second input of which receives the delayed signal of the fields O and M and determines the motion vector, which, having passed the entropy coding block 23, is fed to the input of the multiplexer 24. Next, the digital stream combined in the multiplexer is fed to the input of the buffer storage device BZU 26. Using BZU 26, the control unit for the compression ratio UKS 25, quantizers 7 and 11, the multiplexer 24 is equalized speed digital stream in the communication channel, regardless of the content of the transmitted image. This increases the efficiency of using a communication channel, the throughput of which is limited. Compressed digital stream from the output of the BZU 26 enters the communication channel.

In the receiving (decoding) device, the compressed signal is fed to the input of the buffer storage device BZU 1, in which the original density of the digital stream is restored, that is, the data from the BZU are read unevenly in time.

хранится предыдущее значение сигнала для образования в сумматоре 15 текущего значения сигнала N. Аналогичную функцию выполняет и предсказатель 12 для сигналов М-полей.From the outputs of demultiplexer 2, the encoded image data and motion vectors arrive at the entropy coding blocks 3, 6, 10. Difference and reference signals O, N, and M are fed to the dequantizers Kv ^-1 4, 7 and the blocks of the inverse cosine transform DCT ^-1 5, 8 In block 9, the restoration (interpolation) of the missing samples in the image of N-fields occurs. Then, from the difference signals of N- and M-fields in blocks 11-15, the initial structure of N- and M-fields with PCM is restored. Using blocks 10, 11 and 12 of the signals of the motion vectors, the signals of the M-fields of moving images are formed. In the predictor 14 with a memory device

the previous signal value is stored for the current signal N to be formed in adder 15. A predictor 12 for M-field signals performs a similar function.

When receiving signals of the reference O-field, the switch of the electronic switch is in position 1 - the signal O passes to the output of the circuit. Pulses U _{1 of the} reference field O are received at the control input of EC 17. When receiving macroblocks of M- and N-fields, the EC switch is in position 2. In this case, the output signal is formed by elementwise addition of the inverted DCT ^-1 5, 8 values differences with the predicted macroblock formed from elements of previously decoded images of O- and M-fields using decoded motion vectors in M-fields (blocks 10, 11, 12). From the output of EC 17, the decompressed digital stream from the PCM enters the reverse field permutation block OPP 18, in which the original sequence of fields is restored.

An advantage of the invention is the improvement of image quality with a relatively simpler circuit implementation, which has not only economic, but also social effect, as it can be widely used in many areas of digital television.

Literature

1. Lebedev D.S., Tsukkerman I.I. Television and information theory. - L.: Energy, 1965 .-- 219 p.

2. Digital television. / Krivosheev M.I., Vilenchik L.S. et al. - M .: Communication, 1980. - 260 p.

3. Digital coding of television images. / Zuckerman I.I., Katz B.M. et al. - M.: Radio and Communications, 1983 .-- 239 p.

4. Image transfer methods. Reduction of redundancy. / William Pratt, David D. Sacrison et al. / Translation from English. - M .: Radio and communications, 1983. - 284 p.

5. Trofimov B.E., Kulikovsky OV Digital image transmission. - M.: Communication, 1980 .-- 120 p.

6. W. Pratt. Digital image processing. - M.: Mir Publishing House, 1982, T.1-T.2. - 790 s.

7. Kretzmer (Kretzmer E.R.) Reduce-alphabet representation of television signals, "Conv Rec. IRE", 1956, 4, p. 140.

8. Graham (Graham R.E.). Communication theory applied to television coding "Acta Electronica", 1957/58, N1-2. - c. 333.

9. Anufriev I.K. The use of digital signal processing methods in television. // Means of communication. - 1989, No. 4. - p. 3-6.

10. Smirnov V.M., Sorin V.Ya. Transmission of digital TV signals by two-dimensional coding methods. - Communication technology. Ser. Technique of Television, issue 6, 1987. - P. 31-37.

11. Yaroslavsky. Introduction to digital image processing. - M .: "Soviet Radio", 1979. - 312 p.

12. Bykov R.E. The theoretical foundations of television. - St. Petersburg: Publishing. "Doe", 1990. - 287 p.

13. Levchenko V.N. Satellite television. - “BHV - St. Petersburg”: “Arlit Publishing House”, 1998. - 288 p.

14.N.N.Krasilnikov. Theory of transmission and perception of images. - M .: Radio and communications, 1986. - 246 p.

15. Ptachek M. Digital television. Theory and technology. / Per. from Czech / Ed. L.S. Vilenchik. - M .: Radio and communication. 1990 .-- 528 s.

16. Bezrukov V.N. Development and application of elements of the theory of the transformation of image signals in applied television systems. Author's dissertation for the degree of Doctor of Technical Sciences. - M .: 1996. - 45 p.

17. Zubarev Yu.B., Krivosheev M.I., Krasnoselsky I.N. Digital television broadcasting. - M .: 2001. - 521 p.

18. Television: Textbook for high schools. / Ed. V.E.Jaconia. - M .: Radio and communications, 2004 .-- 615 p.

19. Digital processing of television and computer images. / Ed. Yu.B. Zubarev and V.P. Dvorkovich. - M .: ICSTI. 1997 .-- 255 p.

20. Lokshin B.A. Digital broadcasting: - from studio to viewer. - M .: 2001. - 446 p.

21. Smirnov A.V. The basics of digital television. - M .: 2001. - 224 p.

22. Bryce R. Handbook of Digital Television. Per. from English - Zhukovsky: Era, 2001 .-- 230 p.

23. Digital image conversion. / Ed. Prof. R.E.Bykova. - M .: Hot line - Telecom, 2003 .-- 230 p.

24. Smirnov A.V., Peskin A.E. Digital television: - from theory to practice. - M .: Hot line - Telecom, 2005 .-- 352 p.

25. Vlado Demjanowski. Bible CCTV. / Per. from English - M .: "IS-ES Press", 2003. - 344 p.

26. Petrakov A.V., Lagutin B.C. Teleguard. - M.: Energoizdat. - 1998.372 s.

27. Glassman. K. "MPEG is easy." // Magazine "625", 2000, No. 3. - p. 5-45.

28. Converting standards. Application of technical solutions. // Magazine "625". 2005. No. 7. - p. 79.

29. Glassman K. Videocompensation. // Magazine "625". 1997. No. 7. - p. 60.

30. Balobanov V.G. A method of transmitting and receiving a digital television signal. Copyright certificate No. 1642594. Bull. 14. - 1991.

31. Balobanov V.G. A method of compressing a video signal in digital form. Patent No. 2012157. Bull. No. 8. - 1994.

32. Bezrukov V.N., Balobanov V.G. A method of transmitting additional information on a television channel. Copyright certificate No. 586572. Bull. No. 48. - 1977.

33. Artyushenko V.M., Shelukhin O.I., Afonin M.Yu. Digital compression of video information and sound: a Training manual. / Ed. Artyushenko V.M. - M.: Publishing. Bargain. Corporation "Dashkov & Co." 2003 .-- 426 p. 34. Kataev S.I., Khromoi B.P., Bezrukov V.N., Balobanov V.G. Copyright Certificate No. 301675. Bulletin of inventions No. 14, 1971.

Claims (1)

A method of compressing a digital video signal stream, namely, when converting a video signal in order to reduce the digital stream speed in a communication channel in a preprocessing unit (BPO), the image is divided into blocks of 8 × 8 elements and the odd and even fields in each second frame are interchanged in of the entire video sequence, then form a group of fields consisting of signals of the reference O-fields and two difference signals of N- and M-fields, subject them to discrete cosine transform (DCT), quantization, and entropy coding and multiplexing, and for moving objects in the image of M-fields, instead of the values of the elements of the moving blocks, only their coordinates are transmitted - motion vectors, which are also entropy encoded and multiplexed into a common compressed digital stream, the speed of which, thanks to feedback, consisting of a buffer memory (BZU), the control unit of the compression ratio (UKS), quantizers and multiplexer is made constant regardless of the image structure, characterized in that , in order to eliminate redundancy from TV images and, therefore, reduce the speed of the digital stream in the communication channel, the sequence of fields of the digital video signal with pulse-code modulation are divided into groups consisting of:
O-fields, which are transmitted with intraframe coding and are reference for decoding the remaining fields of the group, providing the ability to decode and play the received TV signal;
M-fields transmitted with inter-frame coding by motion-compensated prediction over the nearest field of an adjacent frame;
N-fields, the transmission of which uses interfield coding with prediction from the previous O- and M-fields, in order to increase the clarity of the image, the odd and even fields in each group of frames are interchanged with the necessary restoration of their sequence in the decoding device, and for additional reduction of the digital stream in the communication channel in N-fields, the samples are thinned out by rows and columns in the matrix of signal values with their subsequent restoration on the receiving side.

Images