JP3948024B2 - Image code transcoder and image code transcoding method - Google Patents

Description

  The present invention provides a transcoding of an image code for converting an original image code obtained by encoding an original image into a reduced image code that represents an image having a code total amount smaller than that of the original image code and corresponding to the code amount. Regarding the coder.

  In recent years, the Motion Picture Experts Group (MPEG) system has become widespread as a system for storing or transmitting moving images and audio. For example, a CD-R (Compact Disc-Recordable), a DVD (Digital Versatile Disc), a hard disk, etc. It is used as a recording format for digital storage media and is also used as a data encoding method for digital television broadcasting. The MPEG system is also used as a data encoding system for streaming moving images and audio.

  As a streaming client that performs streaming using streaming data composed of MPEG signals, in addition to a personal computer that receives streaming data by optical communication at a transmission bit rate of 10 Mbps, ADSL (Asymmetric Digital Subscriber Line) at a transmission bit rate of several Mbps is used. A personal computer that receives streaming data, a W-CDMA (Wideband-Code Division Multiple Access) mobile terminal that receives streaming data through a data communication channel having a transmission bit rate of 384 kbps to 2 Mbps, and the like are assumed. On the other hand, MPEG data produced by a content creator is stored in the streaming server, and this MPEG data is high-quality original MPEG data. If original MPEG data based on conventional television signals such as NTSC and PAL / SECAM are transmitted as they are, a bit rate of about 6 Mbps is required.

  Therefore, in order to transmit streaming data to various streaming clients, it is necessary to convert the original MPEG data into secondary MPEG data having a lower transmission bit rate that is required by the transcoder.

  Conventionally, as a method of reducing the amount of code by a transcoder, (1) a method in which a video signal is decoded by an MPEG decoder and then the decoded video signal is encoded by an MPEG encoder, and (2) a quantization scale is adjusted. (3) A method of reducing the code amount by thinning out the P picture, (4) A method of reducing the code amount by converting the I picture to the P picture, and (5 A method for reducing the code amount by downscaling has been proposed.

  A method of decoding a video signal with an MPEG decoder and then encoding the decoded video signal with an MPEG encoder will be described.

  This method is the most primitive method. After the video signal is decoded by the MPEG decoder shown in FIG. 22, the video signal is encoded by the MPEG encoder shown in FIG.

  Referring to FIG. 22, the MPEG decoder includes a buffer 701, a variable length decoder 702, an inverse quantizer 703, an inverse discrete cosine transformer (IDCT) 704, an adder 705, a past image memory 706, a future image memory 707, a motion. A compensation predictor 708 and a switch 709 are provided. This MPEG decoder receives a video stream 751 and outputs a video signal 752. Further, the inverse quantization characteristic of the inverse quantizer is determined by the quantization scale 753. The quantization characteristic will be described later. Further, since the operation of each part of the MPEG decoder is known, the description thereof is omitted.

  Referring to FIG. 23, the MPEG encoder includes a switch 801, a discrete cosine transformer (DCT) 802, a quantizer 803, an inverse quantizer 804, an inverse discrete cosine transformer (IDCT) 804, an adder 806, and a past image memory. 807, a future image memory 808, a motion detection unit 809, a motion compensation predictor 810, a switch 811, a variable length encoder 813, a buffer 814, and a control unit 815. This MPEG encoder inputs a video signal 752 output from the MPEG decoder as a video signal 851, and outputs a video stream 852. The quantization characteristics of the quantizer 803 and the inverse quantizer are determined by the quantization scale 853. Since the operation of each part of the MPEG encoder is known, its description is omitted.

  In the transcoder in which the MPEG decoder and the MPEG encoder are connected in series, transcoding can be performed by making the bit rate for reading the video stream from the buffer 814 different from the bit rate of the video stream 751. In this case, the quantization scale 853 changes according to the bit rate at which the video stream is read from the buffer 814. Transcoding can also be performed by changing the configuration of a GOP (Group Of Pictures) determined by the MPEG encoder.

  A method of reducing the code amount by adjusting the quantization scale will be described.

  In order to realize this method, for example, a transcoder as shown in FIG. 24 is used. This transcoder includes a buffer 901, a variable length encoder 902, an inverse quantizer 903, a quantizer 904, a multiplexer 905, a variable length encoder 906, a buffer 907, and a control unit 908. Since the buffer 901, variable length decoder 902, and inverse quantizer 903 are the same as those of a known MPEG decoder, description thereof will be omitted. The quantizer 904 re-quantizes the DCT coefficient output from the inverse quantizer 903 according to the transcoded quantization scale 954 output from the control unit 908. The multiplexer 905 multiplexes the numerical value indicating the level of the requantized DCT coefficient output from the quantizer 904 and data other than the DCT coefficient. The variable length encoder 906 performs variable length encoding on the data output from the multiplexer 905. This variable length coding includes two-dimensional Huffman coding of DCT coefficients. The buffer 907 absorbs fluctuations in the frequency of occurrence of variable length codes output from the variable length encoder 906 and outputs a transcoded video stream 952 at a constant bit rate. The control unit 908 determines the value of the transcoded quantization scale 954 according to the occupation ratio of the buffer 907, predetermined rules, and the like. By changing the bit rate at which the transcoded video stream 952 is read from the buffer 907 to be different from that of the original video stream 951, the value of the transcoded quantization scale 954 can be adjusted. Also, the transcoded quantization scale 954 adjusted in this way is generally different from the original quantization scale. Therefore, the original quantization scale 953 is converted to the transcoded quantization scale 954.

  The MPEG decoder for decoding the transcoded video stream 952 is the same as the MPEG decoder shown in FIG. 22, and this MPEG decoder uses the transcoded quantization scale 954 as the quantization scale 753.

  In the transcoder quantizer 904 shown in FIG. 22, regardless of whether the macro block is an intra macro block or a non-intra macro block, 8 × 8 pixels of each DCT block in each macro block are DCT-converted. The 8 × 8 DCT coefficients obtained by the conversion are quantized. As a result of this quantization, a numerical value indicating the amplitude of the DCT coefficient after quantization is obtained. In the MPEG decoder shown in FIG. 22, the quantized DCT coefficient is restored from the numerical value indicating the amplitude of the quantized DCT coefficient regardless of whether the macroblock is an intra macroblock or a non-intra macroblock. Then, the restored 8 × 8 DCT coefficients are subjected to inverse DCT transform to restore 8 × 8 pixels.

The AC coefficient F (m, n) (m, n is an integer satisfying m + n ≠ 0) among the DCT coefficients of the DCT block in the intra macroblock is a quantum having no dead zone in the vicinity of zero amplitude. Quantized by the quantizer. Since the step size S1 of this quantizer differs depending on the component of the AC coefficient, it can be expressed as S1 (m, n). The step size S1 (m, n) is
S1 (m, n) = 2 × intra quantization matrix (m, n) × quantization scale / 32
It is expressed.

The amplitude F1 (m, n) of the DCT coefficient after quantization obtained by the inverse quantizer corresponding to this quantizer is the amplitude F1 (m, n) of the DCT coefficient after quantization output from the quantizer. If the numerical value indicating is QF1 (m, n),
F1 (m, n)
= 2 × QF1 (m, n) × Intra quantization matrix (m, n) × quantization scale / 32
It becomes.

The DCT coefficient F (m, n) (m, n is an integer greater than or equal to 0) of the DCT block in the non-intra macroblock is quantized by a quantizer having a dead zone in the vicinity of zero amplitude. Since the step size S2 of this quantizer varies depending on the DCT coefficient component, it can be expressed as S2 (m, n), but the step size S2 (m, m) is
S2 (m, n) = 2 × quantization scale × non-intra quantization matrix (m, n) / 32
It is expressed.

The amplitude F2 (m, n) of the DCT coefficient after quantization obtained by the inverse quantizer corresponding to this quantizer is the amplitude F2 (m, n) of the DCT coefficient after quantization output from the quantizer. If the numerical value indicating is QF2 (m, n),
F2 (m, n)
= (2 * QF2 (m, n) + k) * non-intra quantization matrix (m, n) * quantization scale / 32
However,
k = sign (QF2 (m, n))
It becomes.

Here, when the value of q_scale_type is 0, the quantization scale of the above two equations is
When the quantization scale = 2 × quantization scale code and the value of q_scale_type is 1,
Quantization scale = function (quantization scale code)
It is expressed. Here, function indicates a non-linear characteristic. FIG. 25 shows an example of the characteristics of the quantization scale code versus the quantization scale.

  The q_scale_type can be determined for each picture, and the quantization scale code can be determined for each macroblock. Therefore, by adjusting the value of q_scale_type for each picture and adjusting the value of the quantization scale code for each macroblock, the quantization scale for each macroblock can be increased. By increasing the quantization scale, the numerical value representing the DCT coefficient of the quantized DCT block can be reduced to reduce the code amount.

  A method of reducing the code amount by thinning out P pictures will be described.

As shown in FIG. 26, the code amount can be reduced by the amount of the thinned out picture by thinning out the picture. Assume that a GOP (Group Of Pictures) having a structure as shown in FIG. When the code amount is slightly reduced, the B ₂ picture, the B ₅ picture, and the B ₈ picture are thinned out as shown in FIG. In this case, for example, the decoder may interpolate B ₂ picture with I ₁ picture, B ₅ picture with P ₄ picture, and B ₈ picture with P ₇ picture. There is no need to re-encode the undecimated pictures. Further, when the code amount is reduced, as shown in FIG. 26C, the B ₂ picture, the B ₃ picture, the B ₅ picture, the B ₆ picture, the B ₈ picture, and the B ₉ picture are thinned out. In this case, in the decoder, for example, B ₂ picture and B ₃ picture are interpolated with I ₁ picture, B ₅ picture and B ₆ picture are interpolated with P ₄ picture, and B ₈ picture and B ₉ picture are interpolated with P ₄ picture. Since it is sufficient to interpolate with ₇ pictures, the transcoder does not need to re-encode the pictures that have not been thinned out. Further, when the code amount is reduced, as shown in FIG. 26 (d), in addition to the B ₂ picture, the B ₃ picture, the B ₅ picture, the B ₆ picture, the B ₈ picture and the B ₉ picture, the P ₄ picture Thin out. In this case, in the decoder, for example, B ₂ picture, B ₃ picture, P ₄ picture and B ₅ picture are interpolated with I ₁ picture, and B ₆ picture, B ₈ picture and B ₉ picture are replaced with P ₇ picture. Complement. However, P ₇ pictures, because it is originally motion compensated prediction from P ₄ picture, the decoder, the P ₄ picture are thinned out, it is impossible to decode the P ₇ picture. Therefore, the transcoder, and motion compensation predicting P ₇ pictures from I ₁ picture, it must encode the prediction error.

  A method for reducing the code amount by converting the I frame into the P frame will be described.

The GOP is composed of an I picture, a P picture, and a B picture, but the code amount per picture is the largest for the I picture, followed by the P picture and the B picture. Therefore, as shown in FIG. 27, the code amount is reduced by converting the I ₁₀ picture of GOP2 into the P ₁₀ picture. For this purpose, the transcoder, and motion compensation predicting P ₁₀ pictures from P ₇ pictures, must encode the prediction error.

FIG. 28 shows an example in which the method of reducing the code amount by thinning out the B picture and the P picture and the method of reducing the code amount by converting the I frame into the P frame are shown in FIG. , B ₂ picture, B ₃ picture, P ₄ picture, B ₅ picture, B ₆ picture, P ₇ picture, B ₈ picture, B ₉ picture, B ₁₁ picture, B ₁₂ picture, P ₁₃ picture, B ₁₄ picture, B ₁₅ pictures, P ₁₆ pictures, B ₁₇ pictures, and B ₁₈ pictures are thinned out, and I ₁₀ pictures are converted into P ₁₀ pictures. In this case, the P ₁₀ picture is predicted by motion compensation from the I ₁ picture which is the closest past picture.

  A method for reducing the code amount by downscaling will be described.

  Reducing the number of pixels included in an image is called downscaling. For example, the number of pixels included in the image is reduced by reducing a frame of 640 (horizontal direction) × 480 (vertical direction) to a frame of 320 (horizontal direction) × 240 (vertical direction). In this case, a collection of 2 (horizontal direction) × 2 (vertical direction) DCT blocks before downscaling is one DCT block after downscaling. The simplest downscaling involves decoding the image code to obtain a decoded image, and re-encoding the decoded image. When re-encoding, the motion vector obtained when the image code is decoded may be reused.

  By the way, streaming is simultaneously performed by many streaming clients. Therefore, the streaming server needs to transmit streaming data to a large number of streaming clients at the same time. There are various types of streaming clients as described above, and accordingly, there are also various bit rates of streaming data. Therefore, the streaming server must perform a large number of transcodings simultaneously in order to simultaneously generate various bit rate streaming data.

  When the primitive transcoding method is used, all the decoding processes and all the encoding processes are necessary, so that a large amount of processes are required.

  Also, for example, transcoding for thinning out P pictures, transcoding for converting I pictures to P pictures, and transcoding for downscaling require motion prediction vectors for each macroblock in order to perform motion compensation prediction. And In some cases, a motion prediction vector before transcoding can be reused in a macroblock after transcoding, but in some cases, it cannot. When the motion prediction vector before transcoding cannot be reused for the macroblock after transcoding, since the generation of the motion prediction vector requires a large amount of computation, such transcoding has a large amount of computation. Therefore, it is almost impossible to perform a large number of these types of transcoding using a CPU in one streaming server in view of the current CPU capability. Transcoding that adjusts the quantization scale requires less computation than transcoding that thins out P pictures, transcoding that converts I pictures to P pictures, and transcoding that performs downscaling. However, even when this transcoding is used, it is very difficult to perform a large number of this kind of transcoding using a CPU in one streaming server in view of the current CPU capability. In particular, since the number of streaming clients that simultaneously perform streaming is expected to increase in the future, this problem is expected to become serious.

  Also, in transcoding that reduces the amount of code by adjusting the quantization scale, all DCT coefficients of all DCT blocks in the same macroblock are uniformly coarsely quantized. to degrade. In transcoding that reduces the amount of code by thinning out P-pictures, the frame rate is lowered, so the motion becomes jerky and the image quality deteriorates. In transcoding that reduces the amount of code by converting an I picture to a P picture, error resistance is weakened, so even if a slight error occurs, the image quality deteriorates. In transcoding that reduces the amount of code by performing downscaling, the screen becomes smaller, so the resolution is lowered and the image quality is degraded.

  The present invention has been made in view of the above problems, and an object of the present invention is to provide a transcoder and a transcoding method that can reduce the amount of calculation and maintain a constant image quality.

According to the present invention, a decoding means for obtaining a frequency component of each encoded block of an image by decoding the first image coding, the coded block per the plurality of when re-encoding a plurality of coded blocks Amplitude reducing means for reducing or zeroing all or part of the non-zero frequency components of each coding block so that the code amount is reduced, and a frequency component whose amplitude is zero by the amplitude reducing means When the amplitude of a certain frequency component is increased to a predetermined non-zero, the total code amount per block to which the frequency component belongs is larger than the total code amount per block to which the frequency component belongs before the increase. If the determination result is affirmative, the amplitude increasing means for increasing the amplitude of the frequency component to the predetermined non-zero value, the amplitude decreasing means, Serial An image code transcoder comprising a recoding unit, the obtaining a second image code to re-encode the frequency component of each block processed by the augmentation means, said amplitude reducing means, all or a single There is provided an image code transcoder characterized by dividing a frequency component of a part by an integer larger than 1, rounding down the quotient to make an integer, and multiplying the integer by the integer.

  In the above-described image code transcoder, measurement means for measuring a reduction rate from the amount of the first image code to the amount of the second image code, and the measured reduction rate become a target reduction rate And a control means for changing the integer.

  According to the present invention, since transcoding is performed by reducing or reducing the DCT coefficient amplitude of each DCT block to zero, it is possible to reduce the amount of computation for transcoding. Transcoding can be performed.

  Further, according to the present invention, the amplitude of the DCT coefficient does not fluctuate so much without adopting a method of simply eliminating the high frequency component of the DCT coefficient of each DCT block. A decoded image corresponding to the bit rate can be obtained from the code.

  The best mode for carrying out the present invention will be described below in detail with reference to the drawings.

  FIG. 1 shows the configuration of a transcoder according to an embodiment of the present invention. FIG. 2 shows a format of an input / output signal of a transcoder according to an embodiment of the present invention. The transcoder shown in FIG. 1 receives the original program stream 151 in the format shown in FIG. 2, transcodes it, and outputs a transcoded program stream 161 in the format shown in FIG. The program stream is defined by the MPEG2 standard. As apparent from FIG. 2, the transcoded program stream 161 is different from the original program stream 151 in that the slices (slices 1 to 6) are shorter, and the other parts are the same. .

  Referring to FIG. 1, a transcoder according to an embodiment of the present invention includes a video PES (Packetized Elementary Stream) detection unit 101, a video PES transcoder 102, a transcoded video PES FIFO 103, a PES counter 104, an original program stream FIFO 105, and a D type. A flip-flop 106, a subtractor 107, a comparator 108, a latch 109, a transcoded video PES counter 110, a comparator 111, a rising edge detection circuit 112, an RS flip-flop 113, and a multiplexer 114 are provided.

  The video PES detection unit 101 mainly detects a video PES in the original program stream 151. The video PES detection unit 101 receives the original program stream 151, starts the PES from the original program stream 151, the video PES enable signal 171 that becomes active when the video PES is detected in the original program stream 151, and the original program stream 151. A PES start code detection signal that becomes active when the code detection signal is detected and a video PES end detection signal 173 that becomes active when the end of the video PES in the original program stream 151 is detected are output.

  The video PES transcoder 102 reduces the amount of image codes per DCT block included in each slice in the video PES. This reduction shortens each slice and shortens each video PES. The video PES shortened in this way is the transcoded video PES 152. The video PES transcoder 102 receives the original program stream 151, the video PES enable signal 171, the video PES end detection signal 173, and the reduction rate instruction signal 176, and receives the transcoded video PES 152, the video PES end detection signal 173, and the transformer. A coded video PES enable signal 174 is output. The transcoded video PES enable signal 174 is a signal that becomes active when the video PES transcoder 102 outputs a valid transcoded video PES 152. A timing diagram of input / output signals of the video PES transcoder 102 is shown in FIG. The video PES enable signal 171 becomes active in the video PES (video PES1, video PES2, and video PES3) portions. The video PES end detection signal 173 becomes active at the end of the video PES (video PES1, video PES2, video PES3). The transcoded video PES 152 differs from the video PES in the original program stream in that each slice (slice 1 ', ... slice 6') is shorter. The transcoded video PES enable signal 174 is a portion including a valid image code in each slice of the transcoded video PES and a portion other than the slice (user data 1, user data 2, user data 3). Become active. A detailed description of the configuration and operation of the video PES transcoder 102 will be given later.

  Returning to FIG. 1, the transcoded video PESFIFO 103 uses only the transcoded video PES enable signal 174 as a write enable (WEN), so that only the valid transcoded video PES and the video PES end detection signal 173 are received. These are temporarily stored and output according to a multiplex control signal (described later) output from the RS flip-flop 113. Therefore, the transcoded video PESFIFO 103 has a function of removing the end of the slice without the image code in the video PES and a function of variably delaying the transcoded video PES.

  The PES counter 104 receives the PES start code detection signal 172, and outputs a PES count 175 indicating the counted value.

  The original program stream FIFO 105 receives an inverted signal of the video PES enable signal 171 as a write enable (WEN) signal, and temporarily stores the original program stream 151 and the PES count 175 according to the inverted signal. Therefore, as shown in FIG. 4, the original program stream FIFO 105 includes a PES count (n) corresponding to a PES other than the video PES (user data PES1, audio PES1, audio PES2) and a PES other than the video PES in the original program stream 151. , N + 3, n + 4) are temporarily stored. The original program stream FIFO 105 outputs temporarily stored data according to a multiplex control signal (described later) output from the RS flip-flop 113. Therefore, the original program stream FIFO 105 has a function of removing the video PES from the original program stream 151 and a function of variably delaying the original program stream 151 from which the video PES has been removed.

  The D-type flip-flop 106, the subtractor 107, the comparator 108, the latch 109, the transcoded video PES counter 110, the comparator 111, the rise detection circuit 112, and the RS flip-flop 113 are connected to the multiplex control signal (the multiplexer 114). And a signal for controlling the read enable (REN) of the transcoded video PESFIFO 103 and the read enable (REN) of the original program stream FIFO 105). These timing diagrams are shown in FIG.

  First, since the REN of the original program stream FIFO 105 is active, the PES count having the value n and the user data PES1 are output from the original program stream FIFO 105. When the user data PES1 is completed, the value of the PES count changes from n to n + 3. Therefore, 3 is output from the subtractor 107, the output of the comparator 108 is HIGH, 3 is latched in the latch 109, The coded video PES counter 110 is loaded with “1”, and the RS flip-flop 113 is reset.

  Since the RS flip-flop 113 is reset, the reading of the user data 1 from the original program stream FIFO 105 is stopped, and the video PES 1 ′ is read from the transcoded video PES FIFO 103. Further, the output of the multiplexer 114 is switched from the user data PES1 to the video PES1 '. At the end of the video PES 1 ′, the video PES end detection signal read from the transcoded video PESFIFO 103 becomes HIGH, so that the count value of the transcoded video PES counter 110 increases from 1 to 2. Subsequently, the video PES 2 ′ is read from the transcoded video PES FIFO 103. At the end of the video PES 2 ′, the video end detection signal read from the transcoded video PESFIFO becomes HIGH again, so that the count value of the transcoded video PES counter 110 increases from 2 to 3.

  When the count value of the transcoded video PES counter 110 becomes 3, this value and the value 3 held by the latch 109 become equal, so the output of the comparator 111 becomes active (HIGH). The rise detection circuit 112 detects a change in the output of the comparator 111 and generates a set signal for the RS flip-flop 113. Therefore, since the RS flip-flop 113 is set at this time, reading of the video PES 2 ′ from the transcoded video PES FIFO 103 is stopped, and the audio PES 1 is read from the original program stream FIFO 105. Further, the output of the multiplexer 114 is switched from the video PES 2 'to the audio PES1.

  When the audio data PES1 is completed, the value of the PES count changes from n + 3 to n + 4. Therefore, although 1 is output from the subtractor 107, the output of the comparator 108 remains LOW. Accordingly, the latch 109 remains latched 3, the count of the transcoded video PES counter 110 remains “3”, and the RS flip-flop 113 remains set. Accordingly, the audio PES2 is read from the original program stream FIFO 105 next.

  Since the subsequent operation is the same as described above, the description thereof is omitted.

  With such an operation, the multiplexer 114 outputs a transcoded program stream 161 having a format as shown in FIG. The bit rate can be reduced by making the read clocks of the transcoded video PES FIFO 103 and the original program stream FIFO 105 slower than these write clocks.

  Next, the video PES transcoder 102 will be described in detail.

  FIG. 6 shows an example of the configuration of the video PES transcoder 102. Referring to FIG. 6, the video PES transcoder 102 includes a slice detector 121, a variable length decoder 122, a DCT coefficient changing unit 123, a multiplexer 124, a variable length encoder 125, a logical product gate 126, and a code amount control unit 127. Prepare.

  The slice detector 121 receives the original program stream 151 and the video PES enable signal 171, detects a video start code and slice from the video PES, and detects the original program stream 151, the video PES enable signal 171, the video start code detection signal 180, and A slice enable signal 189 is output. The video start code detection signal 180 is a signal that becomes active in a portion where the video start code is present in the original program stream 151 output from the slice detection unit 121. The video start code is a start code having a value of 00H to B8H such as a picture start code (00H) and a slice start code (01H to AFH). The slice enable signal 189 is a signal that becomes active in the slice portion of the original program stream 151 output from the slice detection unit 121.

  The variable length decoder 122 receives the original program stream 151, the video PES enable signal 171, the video start code detection signal 180, and the slice enable signal 189, decodes the variable length code in the video PES, and outputs the DCT coefficient of each DCT block. The pre-change decoded data 181 including is output.

  The DCT coefficient changing unit 123 receives the DCT coefficient of each DCT block in the variable length decoded data 181 and outputs all or part of the DCT coefficients of each DCT block with the amplitude reduced or zero. A DCT coefficient whose amplitude is not reduced or made zero is output as it is. Further, the DCT coefficient changing unit 123 changes the degree of reducing or reducing the DCT coefficient to zero according to the value of the coefficient changing signal 188. A method of reducing or zeroing the DCT coefficient performed by the DCT coefficient changing unit 123 will be described in detail later.

  The multiplexer 124 receives the DCT coefficient of each DCT block from the DCT coefficient changing unit 123, receives data other than the DCT coefficient from the variable length decoder 122, multiplexes the input data, and outputs the multiplexed data.

  The variable length encoder 125 performs variable length encoding on the data multiplexed by the multiplexer 124 and outputs the variable length code as the transcoded video PES 152. This variable length coding also includes variable length coding of the DCT coefficients of each DCT block. The variable length encoder 125 also outputs a transcoded video PES enable signal 174 indicating whether each byte of the output transcoded video PES 152 is valid. The transcoded video PES enable signal 174 shown in FIG. 3 is continuously active from the start point to the middle of the slice. This is because the variable length encoder 125 continuously outputs the variable length code in the slice from the beginning of the slice. This is the case when it is generated. Since the DCT coefficient of each DCT block is reduced or zeroed by the DCT coefficient changing unit 123, a variable-length code is not generated from the middle to the end of the slice. The video PES enable signal 174 becomes inactive. Note that the variable length encoder 125 may intermittently generate a variable length code, and in this case, the transcoded video PES enable signal 174 is also intermittently activated.

  The logical product gate 186 takes the logical product of the inverted signal of the image code enable signal 185 and the video PES enable signal 171 and outputs the logical product as the image code disable signal 186.

Referring to FIG. 7, the code amount control unit 127 includes an image code enable time measurement unit 201, an image code disable time measurement unit 202, a reduction rate measurement unit 203, and a coefficient change signal adjustment unit 204. The image code enable time measuring unit 201 receives the image code enable signal 185 and measures the accumulated value T1 of the time during which this signal is active. The image code disable time measuring unit 202 receives the image code disable signal 186 and measures the accumulated value T2 of the time during which this signal is active. The reduction rate measuring unit 203 calculates the measurement reduction rate R by the following equation:
R = T2 / (T1 + T2)
And a measurement reduction rate signal 213 indicating this is output. The coefficient change signal adjustment unit 204 receives the reduction rate instruction signal 176 and the measurement reduction rate signal 213 so that the measurement reduction rate R indicated by the measurement reduction rate signal 213 becomes the target reduction rate indicated by the reduction rate instruction signal 176. Then, the value of the coefficient change signal 188 is changed.

  Next, a method for reducing or zeroing the DCT coefficient performed by the DCT coefficient changing unit 123 will be described in detail.

  The DCT coefficient changing unit 123 changes the input DCT coefficient of each DCT block so that the code amount per DCT block is reduced when re-encoding. As a changing method, for example, the following methods (1) to (7) are used. That is, the DCT coefficient changing unit 123 reduces the amplitude of all or part of the input non-zero DCT coefficient to a smaller amplitude or sets it to 0 by using any one of the methods (1) to (7). Thus, the code amount per DCT block is reduced.

  (1) A low-pass filter is applied to the DCT coefficient of each DCT block to reduce or zero the amplitude of the high-frequency DCT coefficient.

  For example, a two-dimensional low-pass filter having frequency characteristics as shown in FIG. 8 is prepared. The frequency characteristic of the two-dimensional low-pass filter is divided into four frequency regions (region A, region B, region C, and region D). The gain in each frequency domain is 1 (area A), 3/4 (area B), 2/4 (area C), or 1/4 (area D). The two-dimensional low-pass filter multiplies each DCT coefficient by this gain and rounds it down for integerization. When this two-dimensional filter is applied to, for example, the DCT coefficient f (i, j) shown on the left in FIG. 9, a DCT coefficient F (i, j) shown on the right in FIG. 9 is obtained. When this two-dimensional low-pass filter is applied, f (2,3) = 1 of nonzero DCT coefficients f (i, j) changes to F (2,3) = 0, but other nonzero The coefficient f (i, j) does not become zero, but the amplitude decreases.

  The two-dimensionally distributed DCT coefficients are arranged in a one-dimensional DCT coefficient by zigzag scanning as shown in FIG. That is, the DCT coefficients are f (0, 0), f (0, 1), f (1, 0), f (2, 0), f (1, 1), f (0, 2),. . ,. They are arranged one-dimensionally in the order of f (7, 7). These DCT coefficients are coded by pairing the amplitude (level) of the non-zero DCT coefficient and the continuous number of zero DCT coefficients (zero run length) preceding the non-zero DCT coefficient when arranged in this way. It becomes. A variable length code having a length corresponding to the occurrence frequency is assigned to each pair of level and zero run length. The relationship between the level of each pair, the combination of zero run length, and the length of the variable length code is as shown in FIGS. FIG. 11 is for a non-intra block and an intra block with intra_vlc_format = 0, and FIG. 12 is for an intra block with intra_vlc_format = 1. However, FIG. 11 and FIG. 12 show the relationship between the combination of level and zero run length and the length of the variable length code for only a part of all combinations of level and zero run length.

  If the block shown in FIG. 9 is an intra block of intra_vlc_format = 1, if such encoding is performed on f (i, j) before applying the two-dimensional low-pass filter, the left side of FIG. In the two-dimensional matrix, for example, the amplitude (level) of the DCT coefficient f (1, 1) is 1, and the corresponding zero run length is 3 when zero is counted by the zigzag scan of FIG. Is 6 which is a value at level 1 and zero run length 3 in the variable length code length table of FIG.

  Zero run lengths and code lengths corresponding to the amplitudes of other DCT coefficients are also obtained as shown below the two-dimensional matrix on the left side of FIG. Accordingly, the code length per block excluding EOB (End Of Block) is 123 bits.

  On the other hand, when the block shown in FIG. 9 is an intra block of intra_vlc_format = 1, when such encoding is performed on the DCT coefficient F (i, j) after the two-dimensional low-pass filter is applied. The code length is determined as shown below the two-dimensional matrix on the right side of FIG. 9, and the code length per block excluding EOB is 109 bits.

  Accordingly, in the case of an intra block of intra_vlc_format = 1, as shown in FIG. 9, when each non-zero DCT coefficient and the preceding zero run length are two-dimensionally Huffman encoded along the zigzag scan shown in FIG. The code amount per DCT block is 123 bits before the two-dimensional filter is applied, but is reduced to 109 bits after the two-dimensional filter is applied.

(2) The amplitude of the DCT coefficient of each DCT block is divided by an integer greater than 1, and the quotient is rounded down to an integer, and the result of the integerization is multiplied by the integer. That is, when the amplitude before truncation is Amp1, and the amplitude after truncation is Amp2,
Amp2 = int (Amp1 / n) × n
However, n is an integer larger than 1. The value of n may be changed for each frequency of the DCT coefficient. For example, the value of n may be increased as the frequency of the DCT coefficient is higher. In addition, the above method may be performed for all DCT coefficients, but such processing may not be performed for some DCT coefficients.

  For example, as shown in FIG. 13, the frequency region is divided into four frequency regions (region A, region B, region C, and region D). The DCT coefficient in region A is divided by 2, then rounded down to an integer, and then multiplied by 2. Similarly, the DCT coefficient for region B is divided by 4, then rounded down to an integer, then multiplied by 4, and the DCT coefficient for region C is divided by 8, then rounded down to an integer, then multiplied by 8, and The DCT coefficient is divided by 16, then rounded down to an integer, and then multiplied by 16. For example, when this method is applied to the DCT coefficient f (i, j) shown on the left in FIG. 14, the DCT coefficient F (i, j) shown on the right in FIG. 14 is obtained. When this method is applied, f (1,1), f (2,1), f (0,4) f (2,3) and f (6, of the non-zero DCT coefficients f (i, j) 4) changes to zero, but the other non-zero coefficients f (i, j) do not become zero but the amplitude decreases. In the case of an intra block of intra_vlc_format = 1, as shown in FIG. 14, each non-zero DCT coefficient and its preceding zero run length are two-dimensional Huffman encoded along the zigzag scan shown in FIG. The per-code amount is 123 bits before this method is performed, but is reduced to 101 bits after this method is performed.

  (3) When all or part of the amplitudes of the non-zero DCT coefficients of each DCT block are zigzag scanned with the non-zero DCT coefficients, the variable length code corresponding to the set of consecutive zeros preceding the non-zero DCT coefficients The length is reduced or reduced to zero so that the length is shortened by one step or several steps.

  The lengths of the two-dimensional Huffman coding corresponding to the set of non-zero DCT coefficients and the preceding zeros are as shown in FIG. 11 for the non-intra block and the intra block with intra_vlc_format = 0, and intra_vlc_format = 1. The intra block is as shown in FIG. However, FIGS. 11 and 12 show the lengths of the two-dimensional Huffman codes only for some (non-zero coefficients, the number of consecutive zeros preceding them).

  Therefore, as can be seen from FIG. 12, for example, in the case of an intra block of intra_vlc_format = 1, if the amplitude of a DCT coefficient preceded by one zero and having an amplitude of 6 is reduced to 5, the corresponding Huffman code The length can be reduced from 14 bits to 9 bits. That is, in this way, the length of the corresponding Huffman code can be shortened by one step. For example, when such a method is applied to the DCT coefficient f (i, j) of the intra block shown on the left of FIG. 15, the DCT coefficient F (i, j) shown on the right of FIG. 15 is obtained. In the example of FIG. 15, referring to the table of FIG. 12, the amplitude of the DCT coefficient f (i, j) is reduced to the DCT coefficient F (i, j) so that the length of the Huffman code is shortened by one step. Yes. In the case of an intra block of intra_vlc_format = 1, as shown in FIG. 15, each non-zero DCT coefficient and its preceding zero run length are two-dimensional Huffman coded along the zigzag scan shown in FIG. The per-code amount is 123 bits before this method is performed, but is reduced to 90 bits after this method is performed.

  In the example of FIG. 15, the amplitude of each non-zero DCT coefficient is decreased so that the length of the Huffman code is shortened by one step. However, each non-zero DCT coefficient is decreased so that the length of the Huffman code is shortened by several steps. May be reduced or made zero.

(4) In (3), the reduction rate of the amplitude should not exceed a predetermined rate. That is, when the amplitude before change is Amp3 and the amplitude after change is Amp4,
(Amp3-Amp4) / Amp3 <r
However, r is a number greater than 0 and less than 1. The value of r is, for example, 0.9, 0.8, 0.7, 0.6, 0.5, etc. As described above, the deterioration rate of the image quality can be suppressed by preventing the decrease rate of the amplitude from exceeding the predetermined rate.

  FIG. 16 shows the change in the amplitude of the DCT coefficient and the change in the code amount per DCT block when the example shown in FIG. 15 is restricted by r = 0.5. In the case of an intra block of intra_vlc_format = 1, as shown in FIG. 16, each non-zero DCT coefficient and its preceding zero run length are two-dimensional Huffman encoded along the zigzag scan shown in FIG. The code amount per frame is 123 bits before this method is performed in the case of an intra block, but is reduced to 110 bits after this method is performed.

(5) In (3), the amplitude decrease value is made not to exceed a predetermined value. That is, when the amplitude before change is Amp3 and the amplitude after change is Amp4,
(Amp3-Amp4) <diff
However, the value of diff, where diff is an integer exceeding 0, is, for example, 1, 2, 3,. Thus, by preventing the decrease value of the amplitude from exceeding a predetermined value, it is possible to suppress image quality deterioration.

  FIG. 17 shows the change in the amplitude of the DCT coefficient and the change in the code amount per DCT block when the diff = 1 restriction is applied to the example shown in FIG. In the case of an intra block with intra_vlc_format = 1, as shown in FIG. 17, each non-zero DCT coefficient and its preceding zero run length are two-dimensional Huffman coded along the zigzag scan shown in FIG. In the case of an intra block, the per-code amount is 123 bits before this method is performed, but is reduced to 119 bits after this method is performed.

  (6) The amplitude of the last predetermined number s of non-zero DCT coefficients when zigzag scanning the DCT coefficients is set to zero.

  For example, when the value of the predetermined number s is set to 2 and this method is applied to the DCT coefficient shown on the left of FIG. 18, the DCT coefficient shown on the right of FIG. 18 is obtained. In the case of an intra block with intra_vlc_format = 1, as shown in FIG. 18, each non-zero DCT coefficient and its preceding zero run length are two-dimensional Huffman encoded along the zigzag scan shown in FIG. The per-code amount is 123 bits before this method is performed, but is reduced to 75 bits after this method is performed.

  (7) The methods (1) to (6) are combined.

  For example, after applying the method (1) to the DCT coefficient shown on the left of FIG. 14 to obtain the DCT coefficient shown on the right of FIG. 14 (or the left of FIG. 19), the method (4) is further applied. The DCT coefficient shown on the right side of FIG. 19 is obtained. In the case of an intra block of intra_vlc_format = 1, as shown in FIGS. 14 and 19, when each non-zero DCT coefficient and the preceding zero run length are two-dimensionally Huffman encoded along the zigzag scan shown in FIG. The code amount per DCT block is 123 bits before this method is performed, but is reduced to 90 bits after this method is performed.

  (8) In the method of (1) to (3) or (7), in order to reduce the total code amount per DCT block, it was originally a non-zero DCT coefficient, but (1) to (3) or ( All or part of the DCT coefficients that have become zero by the method of 6) are made non-zero DCT coefficients.

  In the example shown in FIG. 9 of the method (1), since f (2, 3) = 1 is changed to F (2, 3) = 0, f (2, 3) and the preceding zero run length are changed. The code length 6 of the code corresponding to (= 2) and the non-zero DCT coefficient f (3, 2) (= 7) after f (2, 3) and the preceding zero run length (= 0) Rather than the sum of code length 7 of the corresponding code (= 13), the code of the corresponding region (the code corresponding to F (3, 2) (= 5) and the corresponding zero run length (= 3)) The code length (= 24) has become longer. In order to solve this problem, the following measures are taken.

  First, by the method (1) to (3) or (7), the amplitude of the non-zero DCT coefficient is reduced or made zero, the DCT block is encoded, and the code amount per DCT block is obtained. Then, if there is a DCT coefficient that has been changed from non-zero to zero by the method of (1), the coefficient is set to non-zero (for example, set to 1 and may be returned to the original amplitude), and the DCT block. And the amount of code per DCT block is obtained. If the latter code amount is smaller, the DCT coefficient changed to zero is set to non-zero.

  When the method (8) is applied to the method (1), the amplitude of F (2,3) becomes 1 as shown on the right side of FIG. In this case, the code amount per DCT block is 97 bits, which is smaller than the code amount of 109 bits in the example shown on the right in FIG.

  When the method (8) is applied to the method (2), the amplitude of F (2, 3) becomes 1 as shown on the right side of FIG. In this case, the code amount per DCT block is 89 bits, which is smaller than the code amount of 101 bits in the example shown on the right in FIG.

  Although no example is shown, the method (8) can be applied to the methods (3) and (6).

  The DCT coefficient changing unit 123 adjusts the reduction ratio of the amplitude of the DCT coefficient or the number of DCT coefficients to be changed according to the value of the coefficient change signal 188, and thereby the reduction ratio of the code amount per each DCT block. Adjust. When the method (1) is used, the change rate of the amplitude of the DCT coefficient can be adjusted by changing the frequency characteristics of the low-pass filter. When the method (2) is used, the rate of change in the amplitude of the DCT coefficient can be adjusted by changing the value of n for each frequency. When the method (3) is used, the number of DCT coefficients to be changed can be adjusted by changing the number of DCT coefficients that reduce the amplitude. Further, the ratio of the amplitude of the DCT coefficient can be adjusted by changing the number of steps of reducing the length of the Huffman code. When the method (4) is used, the number of DCT coefficients to be changed can be adjusted by changing the value of r. When the method (5) is used, the number of DCT coefficients to be changed can be adjusted by changing the value of the predetermined number s. The DCT coefficient changing unit 123 appropriately changes the method to be used or a combination of these methods according to the value of the coefficient changing signal 188, thereby determining the reduction ratio of the amplitude of the DCT coefficient and / or the number of DCT coefficients to be changed. Can be adjusted.

  The configurations of the transcoder shown in FIG. 1, the video PES transcoder shown in FIG. 6, and the code amount control unit shown in FIG. 7 are merely examples when realized by hardware. Further, in the transcoder shown in FIG. 1 and the video PES transcoder shown in FIG. 6, a circuit for delay adjustment in units of several clocks is omitted. The transcoder can be realized by a configuration other than the configuration shown in FIG. In particular, the transcoder can be realized by a computer reading and executing a program for causing the computer to function as a transcoder. Accordingly, the computer can function as a transcoder when the computer reads and executes a program describing a method performed by the transcoder.

It is a block diagram which shows the structure of the transcoder by embodiment of this invention. It is a figure which shows the format of the input / output signal of the transcoder by embodiment of this invention. FIG. 4 is a timing diagram of input / output signals of a video PES transcoder of a transcoder according to an embodiment of the present invention. FIG. 6 is a timing diagram of input signals of an original program stream FIFO of a transcoder according to an embodiment of the present invention. FIG. 6 is a timing diagram of output signals and related signals of a transcoded video PES FIFO and an original program stream FIFO of a transcoder according to an embodiment of the present invention. It is a block diagram which shows the structure of the video PES transcoder of the transcoder by embodiment of this invention. It is a block diagram which shows the structure of the code amount control part of the transcoder by embodiment of this invention. It is a figure for demonstrating the 1st method which the DCT coefficient change part of the transcoder by embodiment of this invention performs. It is a figure which the DCT coefficient change part of the transcoder by embodiment of this invention compares with the DCT coefficient after performing with the DCT coefficient before performing a 1st method with respect to a DCT coefficient. It is a figure which shows the example of the zigzag scan used when carrying out two-dimensional variable length encoding of a DCT coefficient. It is a 1st table | surface which shows the code length of the two-dimensional variable length code by MPEG specification. It is a 2nd table | surface which shows the code length of the two-dimensional variable length code by MPEG specification. It is a figure for demonstrating the 2nd method which the DCT coefficient change part of the transcoder by embodiment of this invention performs. It is a figure which the DCT coefficient change part of the transcoder by embodiment of this invention compares the DCT coefficient after performing with the DCT coefficient before performing the 2nd method with respect to a DCT coefficient. It is a figure which the DCT coefficient change part of the transcoder by embodiment of this invention compares the DCT coefficient after performing with the DCT coefficient before performing a 3rd method with respect to a DCT coefficient. It is a figure which the DCT coefficient change part of the transcoder by embodiment of this invention compares with the DCT coefficient after performing with the DCT coefficient before performing a 4th method with respect to a DCT coefficient. It is a figure which the DCT coefficient change part of the transcoder by embodiment of this invention compares the DCT coefficient after performing with the DCT coefficient before performing the 5th method with respect to a DCT coefficient. It is a figure which the DCT coefficient change part of the transcoder by embodiment of this invention compares the DCT coefficient after performing with the DCT coefficient before performing a 6th method with respect to a DCT coefficient. It is a figure which the DCT coefficient change part of the transcoder by embodiment of this invention compares the DCT coefficient after performing with the DCT coefficient before performing a 7th method with respect to a DCT coefficient. It is a figure which the DCT coefficient change part of the transcoder by embodiment of this invention compares with the DCT coefficient after performing with the DCT coefficient before performing the 1st and 8th method with respect to a DCT coefficient. It is a figure which the DCT coefficient change part of the transcoder by embodiment of this invention compares with the DCT coefficient after performing with the DCT coefficient before performing the 2nd and 7th method with respect to a DCT coefficient. It is a block diagram which shows the structure of the MPEG decoder by a prior art example. It is a block diagram which shows the structure of the MPEG encoder by a prior art example. It is a block diagram which shows the structure of the transcoder by a prior art example. It is a graph which shows the example of the characteristic of the quantization scale code by the MPEG2 standard versus the quantization scale. It is a figure for demonstrating the conventional method of reducing code amount by thinning out a P picture. It is a figure for demonstrating the conventional method of reducing code amount by converting an I frame into a P frame. It is a figure for demonstrating the conventional method of reducing a code amount by thinning out an I picture and a P picture, and also converting an I frame into a P frame.

Explanation of symbols

101 Video PES detection unit 102 Video PES transcoder 103 Transcoded video PESFIFO
104 PES counter 105 Original program stream FIFO
106 D-type flip-flop 107 Subtractor 108 Comparator 109 Latch 110 Transcoded video PES counter 111 Comparator 112 Rise detection circuit 113 RS flip-flop 114 Multiplexer

Claims (6)

Decoding means for decoding the first image code to obtain the frequency component of each encoded block of the image;
As the code amount of the coded block per the plurality of when re-encoding a plurality of coding blocks is reduced, all or reduce the portion of the amplitude or zero non-zero frequency component of each encoded block Amplitude reduction means to
When the amplitude of a certain frequency component among the frequency components whose amplitude has become zero by the amplitude reducing means is increased to a predetermined non-zero, the total code amount per block to which the frequency component belongs is increased, It is determined whether or not the total code amount per block to which the frequency component belongs is smaller, and if the determination result is affirmative, the amplitude that increases the amplitude of the frequency component to the predetermined non-zero value Increase means,
Re-encoding means for re-encoding the frequency component of each block processed by the amplitude reducing means and the amplitude increasing means to obtain a second image code;
The image code transcoder comprising,
The amplitude reducing means divides all or some of the frequency components by an integer greater than 1, rounds the quotient to an integer, and multiplies the integer result by the integer. Coda. The image code transcoder according to claim 1 ,
Measuring means for measuring a reduction rate from the amount of the first image code to the amount of the second image code;
Control means for changing the integer so that the measured reduction rate becomes a target reduction rate;
An image code transcoder. A decoding step of decoding the first image code to obtain a frequency component of each encoded block of the image;
As the code amount of the coded block per the plurality of when re-encoding a plurality of coding blocks is reduced, all or reduce the portion of the amplitude or zero non-zero frequency component of each encoded block An amplitude reduction step to
When the amplitude of a certain frequency component among the frequency components whose amplitude has become zero by the amplitude decreasing step is increased to a predetermined non-zero, the total code amount per block to which the frequency component belongs is increased. It is determined whether or not the total code amount per block to which the frequency component belongs is smaller, and if the determination result is affirmative, the amplitude that increases the amplitude of the frequency component to the predetermined non-zero value Increasing steps,
A re-encoding step of re-encoding the frequency component of each block processed by the amplitude decreasing step and the amplitude increasing step to obtain a second image code;
A transcoding method for image encoding comprising,
In the amplitude reduction step, all or a part of frequency components are divided by an integer larger than 1, the quotient is rounded down to an integer, and the result of the integerization is multiplied by the integer. Coding method. The image code transcoding method according to claim 3 , wherein:
A measurement step of measuring a reduction rate from the amount of the first image code to the amount of the second image code;
A control step of changing the integer so that the measured reduction rate becomes a target reduction rate;
Further image transcoding method code, characterized in that it comprises a. A decoding step of decoding the first image code to obtain a frequency component of each encoded block of the image;
As the code amount of the coded block per the plurality of when re-encoding a plurality of coding blocks is reduced, all or reduce the portion of the amplitude or zero non-zero frequency component of each encoded block An amplitude reduction step to
When the amplitude of a certain frequency component among the frequency components whose amplitude has become zero by the amplitude decreasing step is increased to a predetermined non-zero, the total code amount per block to which the frequency component belongs is increased. It is determined whether or not the total code amount per block to which the frequency component belongs is smaller, and if the determination result is affirmative, the amplitude that increases the amplitude of the frequency component to the predetermined non-zero value Increasing steps,
A re-encoding step of re-encoding the frequency component of each block processed by the amplitude decreasing step and the amplitude increasing step to obtain a second image code;
A transcoding method for image encoding comprising,
In the amplitude reduction step, all or a part of frequency components are divided by an integer larger than 1, the quotient is rounded down to an integer, and the result of the integerization is multiplied by the integer. A program that causes a computer to execute the coding method. In the image code program according to claim 5 ,
The image code transcoding method includes:
A measurement step of measuring a reduction rate from the amount of the first image code to the amount of the second image code;
A control step of changing the integer so that the measured reduction rate becomes a target reduction rate;
A program characterized by further comprising a.

Images