FI89996B - Compression procedure for video signal - Google Patents

Description

399 96 VIDEO SIGNAL COMPRESSION METHOD

The invention relates to a method for compressing a video signal as defined in the preamble of claim 1, in particular for encoding a live video signal into a narrow transmission channel.

The transmission of a video signal typically requires a transmission channel with a throughput of 2 to 70 Mbit / s. When the video signal is compressed by detecting, encoding and transmitting changes or differences in successive image fields, the transmission channel capacity can be reduced between 64 Kbit / s and 2 Mbit / s.

The digital video signal includes image information that is forwarded by pulse code modulation. Such a video signal requires a transmission channel capacity, the size of which is determined by the number of pixels to be transmitted, i.e. pixels, grayscale and the number of image fields to be transmitted per second. The information to be transmitted can be reduced without degrading the image quality 20 by transmitting only a video signal representing the change between successive image fields. To do this, one or another of the image fields must be searched for image areas that have changed relative to the previous image field, and are therefore only transmitted. In addition to the • 25 changed image area information sent, they are • · ·. be provided with an address indicating which area of the image is in question. This makes proper image reconstruction • · ... possible at the receiving end. One such video compression method is disclosed in Finnish Patent Publication 70662.

• 1 ·: Known methods for compressing a live video signal are usually based on a procedure for dividing an image field consisting of e.g. 256 x 256 pixels- »·; 1j1; pixels, to image areas that contain, for example, *. 35 x 8x8 pixels. Using image areas can reduce the number of address bits to be transmitted and thus • ♦ '···' reduce the amount of total information to be transmitted.

* · · · »·· 2 69996

It is an object of the invention to provide a new video compression method for a live video signal in which the changed picture areas of the picture fields are not directly encoded and transmitted to the transmission channel and further to the receiver, but the stored set of previous picture fields is used as a source of estimates of possible changed picture areas.

The method according to the invention is characterized by what is stated in claim 10 1.

In the video signal compression method according to the invention, in particular for encoding a live video signal into a narrow transmission channel, the image fields of the video signal are divided into image areas and the new image field is processed by comparing it According to the invention, a plurality of transmitted picture fields are stored in memory and this set of stored picture fields is used as a source of estimates of possible changed picture areas, at least approximately corresponding corresponding picture areas being searched for, the .25 location of which is indicated as a codeword to the receiver and. based on which codeword estimates the changed image fields among the corresponding image fields stored in the receiver and sets them in the correct image field to reconstruct the image field.

In one embodiment of the method according to the invention, a number of transmitted image fields are stored in an associative memory, from which an estimate of the changed image areas is searched for by means of the contents of the changed image areas and possibly results. 35 one or more memory location addresses from which the estimate * / can be found, which information is forwarded to the receiver * - · ♦ *. The use of associative memory is advantageous in this case, because it is precisely the address information of the estimate that is needed and is sent to the receiver. After all, in conventional direct access memories, the search is carried out using addresses, which in this case 5 would take considerably more time to search for estimates.

In one embodiment of the method according to the invention, the search for the estimate of the changed image area is carried out from the changed image area by means of a simplified image area model. By this is meant that the changed image area is formed into an image area model with a limited but sufficiently correct information content, by which estimates such as this image area model can be quickly and efficiently found in the memory of the stored image fields.

In one embodiment of the method, the image area model is implemented by reducing the number of pixels in the image area to a portion of the original and expressing the tone information contained in each pixel with a limited number of bits. 20 For example, an image area of 8x8 pixels, where the grayscale of each pixel is represented by 6 bits, can be formed into an image area of the same size containing only 4x4 pixels, excluding every other pixel, of which the grayscale of each pixel is indicated by four bits. . Such a rough image area pattern is • close enough to the correct image area when • · '**. the image area is viewed through the human eye. In addition, it should be noted that the est * * * ··· * mate found using the image area model is equally accurate, i.e. it contains 8x8 pixels, and * «30 grayscale is reported as 6 bits / pixel just like the initial

»M

; even the rear changed image area.

The main advantage of the cowardly method of the invention can be considered that the image area of the image field. *: ·. information can be packaged in a much more concise • 35 format than previously known methods.

The method according to the invention achieves, at its most advantageous, up to four times more efficient data transmission than with some known coding methods.

In the following, the invention will be described with reference to the accompanying drawing, in which Fig. 1 schematically shows stored image fields 5 and searching for a changed image area in these image fields; Figure 2 shows a transmitter implemented according to the method; and Figure 3 shows a receiver, respectively.

10 A typical image area, or cluster, has 8x8 pixels, of which the grayscale of each pixel is expressed by six bits. Thus, each pixel is described by 64 grayscale levels. This means that, in theory, there are a total of 15 28x0x6 different image areas. In practice, however, the adjacent pixels are very much similar to each other, and only a limited number of different image areas are found in a real-world image.

The method according to the invention is particularly suitable for data transmission in videoconferencing systems. The camera is usually permanently fixed in place and has a certain part of the room in its field of view. The lighting is relatively constant. People sit and they move relatively little. It can be shown that under these conditions • a large part of the changed image areas tcii • their close estimates can be found in some • * · previous image fields. Thus, the changed image area can be represented by a codeword including location information from a stored field from which the changed image area or its estimate can be found and the image area address here. field. This codeword can be sent over the transmission channel to the receiver instead of ·. * ·: The data of all the pixels in the picture area is encoded and transmitted together with their address bits to the receiver ____: 35. The amount of information to be transmitted can thus be significantly reduced.

'I' The method according to the invention is illustrated in Figure 1. The new image field 1 is processed one image area at a time by comparing it with an image field stored in the reference memory. The image field is processed one image at a time in sequence.

5 The changed image area is observed at a certain point in the image field, such as at A (x ', y'), where x'and y1 represent position coordinates. Thereafter, the image area corresponding to the changed image area or approximately such an image area is searched among the stored image fields 2. A similar image area B is found in this case in the image field N at the position x, y (= certain position coordinates of this image field). The codeword is now formed from the image area number N and the position coordinates x, y, which are sent to the receiver. 15 The set of corresponding picture fields N-1, N, ... N + i is correspondingly stored in the receiver. The image field N is then searched for a similar image area with address information, i.e. location coordinates. x, y and this found image area is placed in the 20 new image fields 1 to be processed, and thus the image field has been reconstructed at the receiver end for this image area. Similarly, a number of changed image areas can be found in a new image field to be transmitted, all of which are searched among the stored image fields 2 25 and correspondingly indicated by a codeword. .1. receiver.

____: The efficiency ... of the method according to the invention can be illustrated by the following example. Suppose I1 ’. that the full picture fields of a live video signal can be stored in the picture memories of both the transmitter and the receiver »1 ·’ · 1 1 receiver. The resolution of the image areas is 256 x 256 pixels and the size of the image areas is 8x8 pixels · / · ♦. When using a standard DCT (Discrete Cosine

Transform) or similar coding method in preferred ‘. Under 35 conditions, compression of lbit / pixel can be achieved.

This means that • · * ”· 1 must transmit 64 bits per image area, in addition to 10 address bits, which • ·» »· 6 89996 totals 74 bits. In the method according to the invention, only a 7-bit codeword is transmitted, by means of which it is indicated from the transmitter to the receiver from which of the 100 stored image fields the desired image area 5 can be found. In addition to this, 10 address bits are needed to indicate in which of the 1024 image areas the desired image area can be found. Thus, only 17 bits are needed to transmit data from the desired image area, which is approximately 4: 1 more efficient than the above-mentioned DCT method.

The transmitter and receiver applying the method according to the invention are shown in Figures 2 and 3. In the transmitter according to Figure 2, the input connection A is an analog / digital level. To the A / D converter 1 and this to the first image memory 2. The transmitter further comprises a second image memory 3. The first and second image memories 2, 3 are connected to the changed image area sensor 4. The sensor 4 is connected to two encoders 5, 6, the first of which the encoder 5 implements the method according to the invention for encoding the changed picture areas and the second encoder 6 for some other suitable method for encoding the picture areas. Both encoders 5, 6 are connected to the selector 7 and further via the output terminal B to a transmission channel, such as a telephone line.

• 25 In addition, the transmitter includes a control unit 8 which • · · .. ·. is connected to different units of the transmitter and with which • ♦ '[[[1 1 transmitter functions are controlled.

• · ... The first encoder 5 comprises a generalization algorithm • · *; ·· [rhythm memory unit 9 and an image field memory unit ** 1 · 30 10. The generalization algorithm memory unit 9 is implemented • ··: preferably as a read-only memory in which the stored image data is permanently stored. region generalization algorithm. The image field memory unit 10 is implemented as an associative or content addressable memory unit (Content Addressable * .35 Memory Unit) in which, for example, 100 image fields are stored.

• · «•« '. ·· 1 The input C of the receiver, Fig. 3, is connected to a divider 11, which is further connected to a first decoding unit 12 and a second decoding unit 13. To the first decoding unit 12 includes an image field memory 14, which is preferably a Random Accessable Memory Unit (RAM-5). The same picture fields are stored in this memory unit as in the transmitter memory unit 10. The second decoder 13 corresponds to the transmitter encoder 6, by which the decoder interprets the transmitted message. Both decoders 12, 13 are connected to the display memory 15 and this is further a digital / analog level. To the D / A converter 16 and further to the output terminal D of the receiver. In addition, the receiver comprises a control unit 17 which is connected to different units of the receiver and by means of which the operation of the different units of the receiver 15 is controlled.

The transmitter and receiver of Figures 2 and 3 operate as follows. The video signal is fed to the transmitter via an A / D converter 1, in which the signal is converted to digital form. The digital video signal is further input to the first image memory 2. The image field then consists of an array of n x m pixels, which is further input to the first image memory 2. The image field of the image memory is divided into image areas consisting of groups of 8x8 pixels. To another image. The previously transmitted reference image • · · • va in the corresponding format stored in the image memory 2 is stored in the memory. with the new image field. From the image memory 2, a digital • · ... video image is input one image area at a time to the ♦ · * · ♦ · * image area sensor 4, to which • · * is also input. *: 30 corresponding image areas from reference memory 3.

• · ·: .r: The found changed image areas are further input in sensor 4 to the generalization algorithm memory unit -: '·. · To 9. At this point, each 8x8 pixel image area contains 6 bits of information per pixel, which • # 35 indicates the grayscale of each pixel. . The memory unit 9 is searched for an image area roughly corresponding to this changed image area, which is the same size as the image area fed to the memory unit, • · »• * · · 8 9 9 9 6 8, but which contains only 4x4 pixels, of which the grayscale of each pixel is represented by four bits. This means that in the generalization algorithm, every other pixel of the 8x8 pixel-5 Iin image area is left unread and the remaining pixels are formed into a 4x4 pixel image area. For grayscale, in the generalization algorithm, the pixels on the 64-step grayscale scale (grayscale is represented by six bits) are described by the nearest points on the 16-10 step grayscale scale (grayscale is represented by four bits). Thus, the image area model shows roughly the essential features of the changed image area.

It should be noted that the representation of the grayscale scale in 15 bits corresponds to 16 different grayscales, while the representation of the grayscale scale in 6 bits contains a total of 64 different grayscales. However, the eye is able to distinguish about 20 shades of gray. Thus, no essential information about the image area is lost by image area modeling performed by the generalization algorithm memory unit 20.

The image area model obtained from the generalization algorithm memory unit 9 is then used as a starting point for searching the image field memory 10 for the estimate '25 for the image area input to the generalization algorithm unit 9.

• The search is performed from the memory using 10 image subject templates • • · .... by comparing it to each stored image • ·. ···. to the image area of the field, i.e. by performing the so-called associative search. As a result of such a search, • · · • · · 30 may find one or more image fields stored in the memory that estimate the said image area. From these, the first or another suitable one is selected, the address information of which, i.e. the image field number N and the estimate address x, y found in the image field, are forwarded as a codeword to the selector 7.

• · ... If there is no * * * suitable estimate for the changed picture area in the picture field memory 10, the picture area • · · • · · · • · 9 J 9 996 is transferred to another encoder 6 which decodes this picture area by some agreed coding method and passes it on to the selector 7. The codewords obtained from the first encoder 5 from the selector 7 as well as the coded picture areas obtained from the second encoder 6 are further transmitted via the output terminal B to the transmission channel.

Alternatively, the transmitter does not use the generalization algorithm memory unit 9, but the identical image area is used to search directly from the image field memory 10 for a similar or approximately similar image area. If one is found, the codeword containing the address information is forwarded just as described above.

Through the receiver input C, the signal from the transmission channel 15 is fed to a splitter 11, from where it is distributed to either the first decoder 12 or the second decoder 13, depending on whether it is a codeword or a fully coded picture area.

The codeword containing the address of the picture area estimate 20 is input to the memory unit 14 of the first decoder. This memory unit 14 contains the same picture field information as the transmitter memory unit 10. However, the address where it is located in this direct access memory is known from the codeword. Address-. Thus, based on 25 teas, an image area estimate is sought, • · 1. . ·. which contains 8x8 pixels and each pixel of which is · · greyscale expressed with a resolution of 6 bits. The found • »... image area estimate is then entered into the display • 1 '·· ** memory at the correct position in the 15 relevant image fields.

• · '. 1:30 The decoded picture area is input in a corresponding manner from the decoder 13 to the picture memory 15 and placed in its correct position in the picture field. Thereafter, the image field completely formed from the image field memory 15 is input • ·: '· 1; to the D / A converter 16, where it is obtained from the output terminal *. 35 via D as a video signal output to the screen.

Image field memory 10 and 14 are filled and • · * ··· 1 is maintained as follows. Image field memory • »· ♦; J 9 9 9 δ ίο 10 is filled by periodically, e.g. once per second, transferring the contents of the image memory 2 to the memory 10 as such instead of the oldest image field there. The image field memory 14 is filled from the memory 15 at regular intervals 5 synchronized with the filling of the transmitter image memory 10 in the same manner to replace the oldest image field therein. In this way, both image field memories 10 and 14 always contain the same image fields and are thus identical.

10 The operation of the receiver and transmitter was shown above in relation to only one image area, but it is clear that all image areas associated with one image area are successively treated and assembled into a single image field 15 at the receiving end in the display memory 15. All the above functions are controlled by the transmitter 8 and in the receiver by means of the control unit 17, preferably by means of the programs stored in the control unit.

The invention has been described above with reference to only one preferred embodiment thereof, but it is clear that the invention can be modified in many ways within the scope of the appended claims.

• · · • «« · · • · • · · · · · · · · · · · · · · · · · · · · · · · i · • · ♦ • · • · · · 9 9 • · r • «· • · · 9 9 9

Claims (4)

3 9 9 96 11 1. A video signal compression method, in particular for encoding a live video signal into a narrow transmission channel, in which the image field of the video signal to be transmitted is divided into image areas and the new image field is processed by comparing it one image area at a time with the previous image memory (3) stored reference field, and only information about the changed image areas is transmitted via the transmission channel to the receiver, characterized in that a plurality of transmitted image fields are stored in the memory (10) and this set of stored image fields is used as a source of estimates of possible changed image areas. at least approximately corresponding altered picture areas, the location of which is indicated as a codeword to the receiver and on the basis of which codeword the corresponding picture fields 20 are searched among the corresponding picture fields 20 stored in the memory (13) of the receiver; stimates and placed in the correct image field to reconstruct the image field. A method according to claim 1, characterized in that the plurality of transmitted image fields are stored in an associative memory (10), »· *; of which changed image areas Estimates are searched using the content of the changed image areas and the results • •. ···. one or more memory locations may be obtained. Kari address where the estimate can be found, which • · · 30 information is forwarded to the receiver. Method according to Claim 1 or 2, characterized in that the search for estimates of the changed image areas is carried out on the changed image area. ϊ from the image area using a simplified image area model. A method according to claim 3, characterized in that the image area model is implemented by *. ** reducing the number of pixels of the image area to a part of the original «· · • • M • · 12 S9996 and expressing the tonal content contained in each pixel. formation with a limited number of bits. • · · • • · · · · · · ··· • · * · · • · · ··· • · • · · · · · · · · · · · · · · • · · • · · · · · · f · · t · · • • · • · · • · • · · · · · · · · · · 3 39996 1. Compression set for a video signal, which can be synced to a home video signal account with 5 small input channels, which can be used for a video signal, with a signal in the picture, with a picture of the picture and with a picture of the picture, bildomräde at gängen jämförs med ett i ett tidigare bildminne (3) lagrat referensbildfält, och 10 self information games lagrade bildfält används säsom en källa för möjl omvandlade bildomrä- 15 dens estimat, varvid ur denna mängd söks fram ätmins- tone ungefär motsvarande omvandlade bildomräden, vilkas plats säsom ett kodord meddelas account mott 13) lagrade motsvarande bildfältens mängd de omvandlade 20 bildfältens estimat söks fram och placeras i rätt bild- Fält för rekonstruering av bildfältet. J [: 2. Förfarande enligt patentkrav 1, k än - ·: ··· netecknat därav, att en mängd sända bildfält. ···. lagras i det associativa minnet (10), varur de omvand- • · ·; 25 estimates of image data are provided in the light of the results obtained and given results • · · • · · * variables may not be available, color estimates are provided, color information is provided • * • · · *. “Vidare mottagaren. • · ·: 30 3. A patent according to claims 1 or 2, which is intended to be included in the invention, is based on an estimate of the average cost of the invention. av en frän det omvandlade bildomrädet förenklad bildom- • ♦ rädes model 1. • · 35 4. A patent according to claim 3, which is incorporated herein by reference in its entirety into a genome of pixels, is shown in Figure 14 89996 as a source and a genome of a third color, in which case 1 or more pixels. • O · 0 0 0 00 0 0 0 0 0 0 0 • 00 mm 00 0 0 0 0 0 • 00 '0 0 • 00 0 0 · 0 0 • 00 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 • 00 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0