DE3926154A1 - SIGNAL PROCESSING SYSTEM - Google Patents

Abstract

A signal processing system for digital signals comprises a quantizer with a non-linear characteristic curve. When sections of a digital signal are coded in different ways, e.g., by interframe and interframe coding, the quality of the decoded signal may vary from one section to another. The system disclosed makes it possible to obtain a decoded signal of uniform quality as well as search mode. This is achieved by using a quantizer characteristic curve for the signal section consisting of a linear portion which is shifted into the field of the characteristic curve in function of the data rate of a preceding signal section, followed by a steadily increasing portion, and by using a virtual buffer memory. The system is suitable for digital signal processing in data-reducing coding.

Description

The invention relates to a signal processing system for digital Signals.

When using digital storage media, e.g. B. Moving image on CD (compact disc) or digital video magnetic tape Recorder, it is desirable that you can also use such modes like "search forward" or "search backward" enables, albeit with reduced image quality. Because in this case information from the previous picture or the previous pictures may be missing, should in certain Intervals can be saved without information the information from directly preceding images can be decoded are, e.g. B. the information for a complete full screen. On these pictures you can z. B. access directly in the search.

In normal operation, the images between the as complete Images coded saved (this corresponds to one "intraframe" coding) only stored in one type of coding, which changes the current image from the previous coded (this corresponds to an "interframe" coding).  To achieve the same subjective image quality in intra- and to achieve interframe encoded images for intraframe encoded images store more data than for interframe encoded.

To a time-varying data rate of a coded digital Signals compared to a storage medium with time constant write or read data rate is to be balanced normally a buffer memory is provided in the encoder or decoder. A data reduction in the coder is e.g. B. by a Data quantization achieved. By controlling the The fill level can correspond to the corresponding quantization characteristic of the buffer storage can be regulated.

By the fonts mentioned below, which is a coding method for video conference transmission and videophone call, it is known that the quantization curve is linear depends on the level of the buffer memory and that at one Scene cut, or when the image content changes significantly between image n-1 and n, the coding for image n + 1 at omitted the recording, the image n with double Data rate intraframe-coded (otherwise interframe) and at Playback as image n + 1 is output as well as the previous one Image n by repeating the image in front of it n-1 is replaced. This discontinuity is caused by the temporal Masking function of the eye covered. The data rate when a scene changes, it remains basically unchanged.

• - CCITT Study Group XV, Document No. 446, 1988, "Description of Reference Model 7" (RM7)
• - CCITT Study Group XV, Temporary Document 10, March 1989, "Reference Model 8 and Action Points" (RM8)

The RM8 codec (codec means: encoder and decoder) and the previous (RM1-RM7) have the disadvantage that  the subjectively perceived image quality between inter and intraframe encoded pictures changes, direct access limited to pictures after a scene change and practically no search is possible if there is no scene change is available.

The invention has for its object a signal processing system for digital signals, e.g. B. to specify a codec subjectively uniform image quality when changing scenes offers and operating modes such as "search forward" or "Search backwards" when using digital storage media enables.

This object is achieved by the specified in claim 1 Features solved. Advantageous developments of the invention are described in the subclaims.

In coding processes in which the coded video signal is based on a medium is stored digitally, you can in periodic Intervals, not only when changing scenes, intraframe-coded Images in the sequence of interframe encoded images insert. Unless there is a scene change, but can do the temporal masking function of the eye here not be exploited.

One solution is to have a very large buffer memory to use the the increased data rate that processed by the intraframe coding of images can.

However, the size of the buffer memory has a direct impact on the minimum decoding delay. The decoder buffer memory must namely have at least the size of the coder buffer memory. When normal playback mode begins, e.g. B. after switching on the device or after a search, the decoder buffer memory is first filled with a certain amount of data before an image is visible at the output of the decoder. The larger the buffer memory, the greater the decoding delay.
A buffer memory that allows a meaningful coding with periodic intra-frame pictures should be in the order of 300 kbit with a data rate of 1.15 Mbit / s.

You also have a simple increase in the buffer memory no control over data rate distribution yet between interframe and intraframe encoded images.

A solution to these problems is provided by a special buffer memory controller, which is the principle of a virtual buffer memory used. The dependence of the quantizer step size the level of the physical buffer memory in the encoder is partially decoupled.

The virtual buffer memory represents the encoded, but not transferred amount of data. Actually controls now the fill level of this virtual buffer memory the quantizer step size.

In the following, "macro block" is understood to mean a block which, for. B. consists of four 8 × 8 DCT luminance blocks arranged in a square and the associated two 8 × 8 DCT chrominance blocks (U and V). The chrominance has half the horizontal and vertical resolution of the luminance.
DCT means: discrete cosine transform.

An embodiment of the invention is described below of the drawings explained. These show in  

Fig. 1 is a block diagram of a coder according to the invention,

Fig. 2 is a known function between a quantizer step size and the filling level of the buffer memory,

Fig. 3 is a function according to the invention between the quantizer and the fill level of the virtual buffer,

Fig. 4 is a function to determine a first one of the parameters for the function of Fig. 3,

Fig. 5 is a function to determine a second one of the parameters for the function of Fig. 3,

Fig. 6 is a function with a low level buffer memory with respect to FIG. 3 modified parameters,

Fig. 7 is a function at elevated buffer fill level changed with respect to FIG. 6 parameters,

Figure 8 is a modified function at a further increased buffer-fill level with respect to FIG. 7 parameters.,

Fig. 9 is a time comparison of the coder input and the encoder output data.

Fig. 1 shows a coder whose input video signal is fed to a circuit 10 which switches between interframe and intraframe coding and a block comparator 172 which checks to what extent an estimated block of pixels corresponds to the corresponding one Original block of pixels matches. A first output signal of circuit 10 is DCT-transformed in circuit 111 , scanned in circuit 121 , weighted in circuit 131 , quantized in circuit 14 , first output signals in circuit 132 inversely weighted, scanned in circuit 122, inverse in circuit 112, DCT- transformed and put together in circuit 19 together with a first output signal of a filter circuit 171 and controlled by a switching signal from circuit 10 to form a block of pixels. The pixel blocks from circuit 19 are buffered in an image memory 18 and fed to the block comparator 172 at the appropriate time. Block comparator 172 computes motion vectors that are encoded in variable word length circuit 152 .
The image signal provided at the output of the block comparator 172 in accordance with the determined motion vector is fed to the filter circuit 171 . The possibly filtered signal is fed to circuit 10 and circuit 19 .

The first output signals of the quantizer 14 are also fed to a circuit 151 , where they (ie DCT coefficients and their addresses) are coded with variable word lengths and are forwarded to the video multiplexer with a physical buffer memory 16 . The output signals of circuit 152 are also fed to this circuit 16 . First output signals of the circuit 16 are then transmitted or stored. The circuit 16 receives as further input signals the output signal of the circuit 10 with the intra / inter decision, a second output signal of the circuit 171 with the filter on / off decision and a second output signal of the quantizer 14 , which indicates the respective quantizer step size to encode the data from circuit 151 accordingly. The quantizer 14 is in turn controlled by a second output signal from the circuit 16 . The circuit 16 and the quantizer 14 with circuit 151 form a control loop.

FIG. 2 shows a linear function of the quantizer step size 21 as a function of the fill level 22 of the buffer memory in the circuit 16 . The quantizer step size 21 can be between Q _min and Q _max . The fill level 22 can be between zero and F _max .

Level 22, quantizer 21, Q _max, Q _min and F _max are shown in Figs. 3 to 8 the corresponding meaning as in Fig. 2.

Fig. 3 shows a function of the quantizer step 31 , which consists of a linear part 33 with the equation Q = Q _min + C₁ × F and a non-linear part 34 with the equation Q = C₂ × F × F + C₃ × F + C₄ , depending on the fill level 32 of a virtual buffer memory. The connection point 35 between the linear part 33 and the non-linear part 34 of the function has the coordinates (F _c , Q _c ). C₁ to C₄ are constants which can be varied by the coder. The virtual buffer memory can be thought of as being included in circuit 16 .

From the virtual buffer memory, a certain number B _x of bits per macroblock or frame is thought to be transferred to the storage medium. B _{x has} the value B _i for intraframe-coded pictures and B _d for interframe-coded pictures. B _i can e.g. B. N _i times the average bit rate B _a bits per macroblock and frame meet. N _i is e.g. B. selected so that a subjectively balanced image quality between intra- and inter-frame coded images.

With a recordable net video bit rate of 1.15 Mbit / s, a frame rate of 25 Hz and a factor N _i = 3, e.g. B. 138 kbit is required for an intraframe-encoded image. The imaginary transmitted data rate B _i bits per macro block will neither significantly increase the fill level of the virtual buffer memory during the intraframe coding of an image, nor will the quantizer step size.

Only a value smaller than B _{a is} permissible for the bit rate B _d . B _d is defined as

B _d = (N _fr × B _a - B _i ) / (N _fr - 1),

where N is coded _for interframe-the period of intraframe encoded pictures within the.
With a recordable net video bit rate of 1.15 Mbit / s, a frame rate of 25 Hz, N _fr = 30 and a use of 138 kbit for an intra-frame coded picture, e.g. B. available on average 46 kbit for each picture and about 42.8 kbit for an interframe-coded picture. The virtual buffer memory has e.g. B. a size of F _max = 92 kbit. Then the physical buffer memory must have a minimum size of given values

F _phmax = F _max + 138 kbit - 46 kbit = 184 kbit

to have.

The control loop between quantizer 14 with circuit 151 and the buffer memory in circuit 16 is advantageously designed such that the fill level 32 of the buffer memory after N _fr -1 interframe-coded pictures is so low that the data of the next intraframe-coded picture in the virtual buffer memory can be saved.

The linear part 33 of the function of the quantizer step size 31 advantageously enables the buffer memory to reach a certain fill level without a significant change in the quantizer step size and thus in the image quality. The advantageously progressively increasing part 34 nevertheless protects the buffer store against overflow. Basically, the non-linear part 34 of the characteristic curve forms a quadratic function.

The minimum quantizer step size Q _min has z. B. the value 4/2048 and the maximum quantizer step size Q _max the value 64/2048.
By using a virtual buffer memory, the coder can advantageously use the same characteristic curve for intra- and interframe-coded pictures, a good picture quality being achievable.

The function of the quantizer step size 33 and 34 can advantageously be adapted to the characteristics of the respective image content by varying the constants C₁ to C₄ and thus shifting the connection point 35 . In order to determine the values from C₁ to C₄ for the current image k, the average buffer fill level F _{a, k-kx} and the average quantizer step width Q _{a, k-kx} for the image k-kx are measured in each case.
In order to achieve a further increased image quality, an optimal characteristic curve for the intra- and inter-mode can advantageously be formed.
In intra mode, kx = N _fr , otherwise kx = 1 or kx = 2 if the last picture k-1 was encoded in intra mode.

The coordinates (F _{c, k} , Q _c ) of the connection point 35 for image k are determined from FIG. 4 from the average virtual buffer fill level F _{a, k-kx of} a previous image k-kx, which was coded in the same mode. The value Q _c advantageously remains constant.

FIG. 5 shows a function 56 of an additional offset Q _{o, k} for the quantizer step size 31 of the current image k as a function of the average quantizer step size Q _{a, k-kx of} the previous image k-kx.

The aim of an adaptation of the quantizer step size functions 33 and 34 is to allow a greater variation of the buffer memory fill level F without a substantial change in the quantizer step size Q, if an image can presumably be coded with a small quantizer step size Q _{a, k} which increases the image quality due to the characteristics of the preceding one.

If the virtual buffer storage level F _a nevertheless rises significantly, the quantizer step size Q must be increased, advantageously by shortening the linear part 33 of the function.

Fig. 6 shows a target from the coder Quantisiererschrittweiten- function.

FIG. 7 shows a function changed from FIG. 6 with a shortened linear part 73 with an increased mean buffer storage level F _{a, k-kx} .

If due to the characteristics of the image k-kx, e.g. B. by fine image details in the entire image, the buffer level is already significantly increased, an offset Q _{o is added} to the quantizer step size Q _c and the minimum quantizer step size Q _min . If the buffer level F in an image is already increased and the quantization is therefore greater, very fine quantization is advantageously not permitted in this image.

FIG. 8 shows a function according to FIG. 7 with an additional offset Q _o for the quantizer step width Q _c or Q _min . Due to the offset Q _o , the function advantageously remains non-linear and limits variations in the quantizer step width Q in the image.

Because the permissible fill level F _{phmax of} the physical buffer memory is greater than the permissible fill level F _{max of} the virtual buffer memory, it is ensured on the one hand that the physical buffer memory does not overflow in the event of a sudden, unpredictably high increase in data rates, and on the other hand the quantizer step size Q increases as the virtual buffer fill level increases early in the vicinity of F _max, thus avoiding a clearly visible jump in image quality. In addition, an unpredictable, strong data rate increase will advantageously only occur when there are scene changes, where it is more tolerated by the viewer (masking function of the eye).

The maximum buffer memory delay T _max results in T _max = F _phmax / R. For F _phmax = 184 kbit and R = 1.15 Mbit / s, the maximum buffer memory delay T _{max is} therefore 0.16 s. This corresponds to the time for 4 pictures at a frame rate of 25 Hz.

Fig. 9 is an example of the approximate temporal relationship between the image data at the encoder input and the coded data at the input of the buffer memory is in the circuit 16.
Intra-frame encoded pictures are labeled i, interframe-encoded pictures are labeled d. The distribution of data rates between I and D images must be such that, as two back face in the first time after N i _fr images images.

Claims (9)

1. Signal processing system for digital signals, in particular video and / or audio signals, which are digitally stored and / or stored on a storage medium, the digital signals consisting of continuous sequences, each of which begins with an intraframe-coded signal section, the a number of interframe-coded signal sections follows, the mutual time intervals of the intraframe-coded signal sections from the signal content, e.g. B. scene changes are essentially independent, characterized in that the quantizer control characteristic ( 33 and 34 ), hereinafter referred to as control characteristic, for each signal section by the level (F) of a buffer (in 16 ) in its step size (Q ) adjustable quantizer is non-linear. 2. System according to claim 1, characterized in that the control characteristic ( 33 and 34 ) consists of a substantially linear part ( 33 ) and an adjoining progressively increasing ( 34 ) part. 3. System according to claim 1 or 2, characterized in that the buffer (in 16 ) contains a virtual buffer (in 16 ), which has a smaller storage capacity than the physical. 4. System according to claim 3, characterized in that at a low and at a medium data rate during a previous signal section or several previous signal sections for the current signal section, the minimum quantizer step width (Q _min ) given by the control characteristic curve remains essentially the same and the connection point ( 35 ) with the coordinates (F _c , Q _c ) between the essentially linear ( 33 ) and the progressively increasing ( 34 ) part of the control characteristic curve ( 33 and 34 ) with a substantially equal quantizer step size (Q _c ) with an increasing data rate of shifts a higher to a lower fill level (F _c ) of the virtual buffer (in 16 ) (to the left in FIG. 3). 5. System according to claim 3 or 4, characterized in that at a high data rate during a previous signal section or several previous signal sections for the current signal section of the substantially linear part ( 33 ) of the control characteristic ( 33 and 34 ) with the connection point ( 35 ) to the progressively increasing ( 34 ) part of the control characteristic ( 33 and 34 ) with increasing data rate essentially parallel in the direction of higher quantizer step width (Q) (upwards in Fig. 3). 6. System according to one or more of claims 1 to 5, characterized in that in each case a different control characteristic ( 33 and 34 ) is formed for interframe or intraframe coded signals. 7. System according to one or more of claims 1 to 6, characterized in that in signal sections which are intraframe-coded, an increased amount of data compared to that in inter-frame-coded data sections, which is the average amount of data of a signal section multiplied by a constant factor corresponds, is read out from the virtual buffer (in 16 ). 8. A code circuit for a system according to one or more of the preceding claims, which is provided with a first circuit which interframe-coded first signal sections and intraframe-coded second signal sections, with a downstream quantizer, which by the during a previous signal section or more previous Signal sections determined fill level (F) of a buffer in its step size (Q) can be controlled in accordance with the control characteristic ( 33 and 34 ) and with a downstream second circuit which codes the respective quantizer step size (Q) and adds the signal content of the buffer. 9. Decoder circuit for a system according to one or more of claims 1 to 7, the stored output signals the code circuit with normal and / or can decode at increased speed and the from the stored output signals of the code circuit in a first circuit the quantizer step size (Q) decoded and in a second circuit with Using this quantizer step size (Q) the original Signal sections before coding accordingly interframe or intraframe decoded, while during decoding with increased speed (search) only the intraframe-coded signal parts with the  corresponding quantization step size (Q) decoded will.