KR100386374B1 - A transcoder - Google Patents

Abstract

The transcoder for video signal conversion between the first and second encoding schemes employing motion compensation includes a decoder 28 for decoding the received data stream encoded according to the first code and the scheme and a second encoding of the data stream from the decoder. Encoder 30 for encoding into a data stream according to the plan. The decoder 28 extracts motion vectors from the received data stream and passes them through the data stream of the encoder to avoid recalculation of the motion vectors. Drift compensation means 52 to 64; 70 to 78 may be provided as means for compensating for any resulting drift after one image frame.

Description

Transcoder {A TRANSCODER}

The present invention relates to a transcoder for signal conversion between a first encoding method and a second encoding method. The invention is particularly suitable for the conversion of video signals.

There is often a need to transmit moving picture television over long distances via a transmission link. When transmitted in digital form, which is expensive to transmit and requires a high bandwidth link, broadcast-characterized televisions require more than 100 Mbit / s. An acceptable degradation of picture quality can be introduced to reduce the content of the transmitted information. Additionally or alternatively, a compression coding technique can be used, which has the advantage of a high level of spatial and temporal margin in the video signal being encoded. So, for example, in videoconferencing applications it is possible to compress up to a few hundred kbit / s bit rate, while video-phone-level images with sound can be compressed to as much as 64 kbit / s equivalent to a single telephone line.

The redundant reduction technique assumes that there is a spatial and / or temporal correlation between adjacent pixels or blocks of adjacent pixels. The details of this correlation as well as the difference between the hypothesis and the actual pixel or block are encoded. Each image frame that is typically encoded consists of an array of pixels (pixels) that are divided into blocks of N × M pixels.

Predictive encoding uses the assumption that a value within a frame is associated with several adjacent values in the same or another frame, so that the value can be calculated at the receiver instead of being transmitted. In this case, it is only necessary to transmit the prediction error that may occur due to this assumption. For example, the first pixel of a frame may be transmitted exactly while each subsequent pixel is transmitted as a difference from the preceding pixel. In more complex methods, the prediction may be a combination of multiple pixels.

Transform encoding takes advantage of the correlation of pixel sizes within a frame by converting pixel sizes to other sets of values, many of which can be encoded using a relatively small number of bits. The most common form of transform coding is Discrete Cosine Transform (DCT). One block of N × M pixels is transformed into an array of transform coefficients of N × M. The resulting coefficient array is quantized by dividing each coefficient by a variable quantization factor. The quantized coefficients may be encoded with variable length codes, such as, for example, Huffman codes.

Another encoding technique is motion compensation, in which one picture is divided into blocks of pixels and each block of the current frame is compared with the corresponding block of the reference frame, where the reference frame is before or after. It may be a frame. Each block of the current frame is then compared with the area shifted from that block to identify the reference frame area most similar to the block.

The vector difference in position between the identified area and the block in question is called a motion vector, and is used to shift the identified area of the reference frame to the position of the relevant block in the current frame. Moving vectors are generated for all blocks of the current frame and they are used to derive the predicted frame from the reference frame. The difference between the current frame and the predicted frame is on average smaller than the difference between the current frame and the reference frame and can be encoded using a lower bit rate. A decoder having a reference frame already stored can reproduce the current frame using the motion vector and the difference value. The signal can be encoded using the coding techniques already mentioned separately or in combination.

It is sometimes desirable to employ a transcoder capable of accepting a received data stream encoded according to the first method and outputting a data stream encoded according to the second encoding method. If it had a decoder operated according to the second encoding method such a transcoder would allow transmission and reception encoded according to the first encoding method without modifying the original encoder. For example, the transcoder is capable of transmitting 64 kbit / s of video signals from an ISDN video terminal in accordance with ITU-T standard H.261 to 32 kbit / s for transmission over a digital European wireless telephone network (DECT). It can be used to convert to the signal of s.

The existing transcoder decodes a video signal encoded according to the first encoding method into an uncompressed video signal encoded by an encoder according to the second encoding method to output a new compressed data stream. When a full decoding operation is performed to reconstruct the original video signal, the video signal is encoded to provide a newly encoded data stream according to the second encoding method. In the case of an encoding method including motion compensation, a new motion vector must be generated for a signal encoded according to a new format, which takes up a large proportion of the processing time of a conventional transcoder.

According to the present invention, a transcoder comprising a decoder for decoding a video signal encoded according to a first encoding method employing a motion compensation technique, and an encoder for encoding the decoded video signal according to a second encoding method, And a decoder extracting motion compensation information from the video signal and transferring the motion compensation information to the encoded video signal output from the encoder.

The motion compensation information can be transmitted unaltered from the decoder to the output side of the encoder, thereby reducing the processing required by the transcoder. Alternatively, the motion compensation information may be modified according to the compression ratios of the two encoding methods.

The decoders and encoders of the transcoder can operate at non-clocked clock rates, and the buffer capacity of the main data path can be reduced compared to conventional transcoders.

The size of the motion compensation block of the two encoding methods does not need to be the same. If the resolutions of the two methods are the same, the motion vector can be transmitted without changing, and if the block size is changed, some of the motion vectors will need to be properly duplicated or discarded. For example, if the motion compensation block size of the first encoding method is 16 × 16 pixels and the motion compensation block size of the second encoding method is 8 × 8 pixels, then the motion vector for one 16 × 16 block is determined by the method of the second method. It can be used for four 8 × 8 blocks arranged in common. If the block size of the first encoding method is 8x8 pixels and the block size of the second encoding method is 16x16 pixels, there is an excess in the motion vector for the second encoding method. So, for example, the average of four related moving vectors can be calculated and used.

If the resolution of the input image differs from that of the output image, the vector is scaled appropriately. For example, if the motion compensation block of both encoding methods is 16 × 16 pixels, the input resolution is 352 × 288 pixels, and the output resolution is 176 × 144 pixels, the vector input to the transcoder is transferred to the encoded video signal. Before it is split into two. The complexity of the calculation is small compared to conventional motion vector estimation.

If predictive coding is used, inconsistencies may occur between data delivered from the original encoder and data decoded at the destination decoder due to transmission errors. Therefore, the decoded image formed at the final decoder will have a defect that increases over time unless unpredicted coding is sometimes employed to restore the integrity of the decoded signal. However, the original encoder may not employ unpredictable encoding often enough so that an acceptable decoded image is built at the destination encoder if a transcoder that is not recognized by the original encoder is employed. This will cause drift between the video signal received by the transcoder and the video signal output from the transcoder.

Therefore, it is advantageous for the transcoder to include drift compensation means. The drift compensation means preferably comprises means for forming a signal indicative of drift between the output of the decoder and the output of the encoder, means for storing the drift signal, and means for adding the drift signal to the output of the decoder. .

In a preferred embodiment, the drift signal is preferably added to the output of the decoder after a delay of one picture period. This configuration compensates for drift without introducing delay in the coded signal.

The drift signal is dependent on a threshold and added to the output of the decoder only when it exceeds the threshold.

Advantageously, said decoder comprises a dequantizer and said encoder comprises a quantizer having a different quantization step size than said dequantizer.

According to a second aspect of the present invention, a transcoder comprising a decoder for decoding a signal encoded according to a first encoding method, and an encoder for encoding the decoded signal according to a second encoding method comprises: an output from the decoder; Compensation means for compensating for drift between inputs to the encoder.

The invention will be described in detail below in an illustrative manner with reference to the accompanying drawings.

1 is a schematic diagram of an existing transcoder;

2 is a schematic diagram of a transcoder according to a first embodiment of the present invention;

3 is a schematic diagram of a transcoder according to a second embodiment of the present invention;

4 is a schematic diagram of a transcoder according to a third embodiment of the present invention;

5 is a schematic diagram of a transcoder according to another embodiment of the present invention; And

6 is an application example of a transcoder according to the present invention.

As shown in FIG. 1, it can be seen that the existing transcoder includes a decoder 28 and an encoder 30. The decoder 28 is a variable that receives a video signal encoded according to a first encoding method employing motion compensation and DCT coding of a difference signal, that is, a 64 kbit / s signal conforming to the CCITT H.261 standard. A length decoder 2. The decoder 2 detects the received data and converts it into quantized DCT coefficients, quantization indexes, and motion vectors. The DCT coefficients are passed to an inverse quantizer 4 and an inverse DCT processor 6 which converts the DCT coefficients to pixel difference values.

The motion vector passes through a motion compensator 8 that computes the address of the predicted pixel block in the previous frame. This block is recovered from the previous frame store 12 and added to the output of the inverse DCT processor 6 to generate a decoded data stream of the current block in the adder 10. The decoded data stream is stored in the previous frame store 12 for reference of the next frame.

The decoded data stream also passes through the encoder 30 of the transcoder, and the motion estimator 14 searches the previous frame buffer 16 to find an offset block of pixels, very similar to the current block. The motion vector of the most suitable block is calculated and the block is recovered from the previous frame buffer 16 and subtracted from the decoded data stream by means 15 to form a differential signal. The differential signal for the current block is converted into the frequency domain by the DCT processor 18. The generated frequency coefficient is quantized in quantizer 20 having a step size suitable for the desired bit rate at the output of the transcoder. The variable length coder 22 converts the output of the quantizer 20 and the motion vector from the motion estimator 14 into variable length codes and outputs the data in a new format.

The encoder 30 of the transcoder also includes a local decoder that includes an inverse quantizer 24 and an inverse DCT processor 26. The inverse DCT processor 26 and the outputs of the motion estimator and compensator 14 are input to an adder 27 to generate an updated prediction frame stored in the previous frame store 16.

A transcoder according to one embodiment of the present invention is shown in FIG. The transcoder comprises a variable length decoder 40 for decoding an incoming signal encoded according to CCITT standard H.261, for example at 64 kbit / s according to the first format. The decoder 40 detects the variable length code and converts it into a DCT coefficient and a motion vector. The motion vector is passed through the transcoder without further processing as indicated by the symbol 42. Such transcoders are suitable for coding methods having the same picture resolution, transform block size, and motion compensation block size. The DCT coefficients are input to inverse quantizer 44, and the resulting data is quantized by quantizer 46 in different quantization step sizes suitable for an output format, for example 32 kbit / s. The new DCT coefficients are encoded by variable length coder 48 and recombined with unmodified motion vector 42 in multiplexer 50.

The decoding is performed directly on the encoded data from the transmission system without buffering and there is little delay to the output of the multiplexer 50, because the process has a short latency. There may be only one buffer that may be needed between the multiplexer and the output side of the transcoder, which will do the same two functions as in a normal encoder. In other words, it is to smooth the coded data and provide additional control of the quantizer 46. However, the smoothing required only covers any locally limited change in output rate due to the discrete nature of quantizer 46, so that this buffer (not shown) and its delay are orders of magnitude. Or smaller than that of a conventional encoder.

3 illustrates a transcoder according to a second embodiment of the present invention, which has a different image resolution, for example, the first encoding method has a resolution of 388 × 288 pixels and the second encoding method is 176 × 144. It is suitable for an encoding method having a pixel resolution. The motion vector scaler 45 scales the incoming motion vector by 0.5 times, and the generated motion vector 42a is multiplexed with the output of the VLC as described with reference to FIG.

Quantizer 46 introduces, at the output of the transcoder, a quantization error that was not predicted at the original encoder. This error or drift will accumulate over time if this drift is not compensated for. According to a second aspect of the invention, the correction is applied directly at the transcoder itself without a change in the path of the motion vector 42. This concept recognizes that the error between the original encoder (not shown) and the final decoder (not shown) will be introduced by requantizing the transform coefficient as soon as possible, dispatching the encoded data to the final decoder, It is to calculate the error introduced and try to correct it at the next opportunity. Of course, at that time a new error will be introduced by requantization of the current coefficient. Also, each correction cannot be perfect due to the discrete nature of the two quantization laws. Thus, the transcoder attempts to calibrate continuously to catch up with the previous error caused.

The transcoder according to FIG. 3 comprises such drift compensation means. The output from the inverse quantizer 44 is fed to an inverse DCT processor 52 and the output of the quantizer 46 is fed to an inverse DCT processor 54. The resulting pixel difference values are output to frame storages 60 and 62, respectively, via adders 56 and 58, respectively, the contents of which are stored in the original motion vector 42 and the scaled motion vector 42a. Rewarded by each. After a delay of one frame, the contents of the frame store 60, 62 are subtracted to form a drift signal between the received data (stored in the frame store 60) and the transmitted data (stored in the frame store 62). 63). This drift signal is converted back into the frequency domain by the DCT processor 64 and added to the output of the inverse quantizer 44 by the adder 65. The drift is compensated after the delay of one frame. The motion compensated content of the frame store 60. 62 also forms a second input of the adders 56, 58, respectively.

The error or drift is obtained by complete coding into the pixel region of both the original coded data and the bitrate reduced version and the formation of the difference. This is then converted and returned to the main path immediately before the quantizer and added. The reconstructed video signal used to form the error is taken from the output of the picture store, with one picture delayed with respect to the main path. Figure 4 shows an alternative embodiment of a transcoder according to the present invention, which is suitable for an encoding method having the same picture resolution, transform block size and motion compensation block size. Similar elements are denoted by the same reference numerals. In this embodiment, the drift signal is formed in the frequency domain as opposed to the pixel region. The outputs of inverse quantizer 44 and quantizer 46 are input to subtractor 70 to form a drift signal between the DCT coefficients of the received signal and the DCT coefficients of the transmitted signal. This drift signal is converted into the pixel region by the inverse DCT processor 72. The output from inverse DCT processor 72 forms one input to adder 74, and the output of the adder is passed to frame store 76. After the delay of one frame as before, the contents of the frame store are retransmitted to the frequency domain by the DCT processor 78. The contents of the frame store also form a second input to the adder 74.

5 shows another embodiment of a transcoder according to the present invention, which is suitable for an encoding method that does not employ a motion compensation technique. Such transcoder includes drift compensation means and is suitable for the use of other signals as well as video signals. As before, similar elements are denoted by the same reference numerals. The transcoder of FIG. 5 operates in the same manner as shown in FIG. When no movement compensation is involved, the drift compensation can be calculated completely in the transform domain.

The present invention works equally well for the application of so-called Variable Bit Rate (VBR) systems when the reduced rate is not constant. An example of particular relevance to packet video is the connection between a terminal connected to an ISDN and another terminal on a LAN, as shown in FIG. The video compressed at a constant rate from the ISDN terminal is delivered to the LAN by the gateway transcoder 80 unchanged when the LAN traffic is sufficiently small. However, during congestion on the LAN, the gateway transcoder 80 may reduce the video data rate required on the LAN. There is no need for a control mechanism to return to the original encoder 84. There is no potential problem of transmission delay time to long range encoders anywhere in the world and potential problems of response time by long range encoders. Moreover, the transcoder is capable of ratio conversion to almost any size and can perform any instant rate conversion compared to steps in the 20 millisecond range possible, for example in discrete 64 kbit / s or H.320 / H.221. have.

Another example fixed in variable bit rate conversion may occur in networks such as mobile applications, and the Automatic Repeat reQuest (ARQ) mechanism dynamically reduces the effective throughput.

In this example of variable bit rate application the transcoder is typically inactive if the quantization indices of dequantizer 44 and quantizer 46 are equal. The picture quality is therefore not impaired by the transcoder. Bit rate reduction of the transcoder 80 only occurs for a relatively short time to mitigate transmission problems. Temporary degradation of picture quality is much more desirable than the usual effect of data loss on prediction algorithms.

The transcoder according to the present invention is also suitable for converting constant bit rate data from a terminal, for example, terminal 86 of a LAN, into VBR data for transmission via an ATM network.

The coded data from the transcoder of the present invention may be transmitted to a decoder operating at the transmitted data rate, or to a decoder operating at a higher data rate, e.g., the data rate of the original encoder.

Claims (9)

A transcoder comprising a decoder for decoding a video signal encoded according to a first encoding method employing a motion compensation technique, and an encoder for encoding the decoded video signal according to a second encoding method. The decoder extracts motion compensation information from the video signal and passes the motion compensation information from the encoder to the coded video signal output side. The method of claim 1, And further comprising drift compensation means. The method of claim 2, The drift compensation means, Means for forming a signal indicative of a drift between an output of the decoder and an output of the encoder; Means for storing the drift signal; And Means for adding the drift signal to an output of the decoder. The method of claim 3, wherein And the drift signal is added to the output of the decoder after a delay. The method of claim 4, wherein And the drift signal is added to the output of the decoder after one picture period delay. The method according to claim 4 or 5, The drift signal forming means forms a drift signal in the frequency domain, Means for converting the drift signal into a pixel region; And And means for converting the drift signal back into the frequency domain. The method according to claim 4 or 5, The drift compensation means includes means for storing the output of the decoder and the output of the encoder in the pixel region and for forming a drift signal in the pixel region, and means for converting the drift signal back into the frequency domain. . The method according to any one of claims 3 to 5, The drift signal is added to the output of the decoder only when the drift signal exceeds a threshold. The method according to any one of claims 1 to 5, The decoder comprises an inverse quantizer and the encoder comprises a quantizer having a different quantization step size than the inverse quantizer.