JP2007135101A - Rate converting apparatus for moving picture - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To provide a rate converting apparatus for moving picture which outputs the moving picture to be input by being encoded into an optional transmission bit rate by converting it into another optional transmission bit rate. <P>SOLUTION: The encoded moving picture is input in a variable length decoding portion 1 by a frame unit. The variable length decoding portion 1 variable-length-decodes the moving picture. A target quantization parameter converting portion 2 converts the quantization parameter QP extracted by the variable length decoding portion 1 into a target value (target quantization parameter) QPt of the quantization parameter after a rate conversion. A conversion system selecting portion 3 selects either "requantization system" or "reencoding system" as a rate conversion system. The requantization portion 5 rate-converts the moving picture in a process to reverse-quantize and requantize it. The reencoding portion 4 rate-converts the moving picture in the process to decode and reencode it until a pixel level. <P>COPYRIGHT: (C)2007,JPO&INPIT

Description

  The present invention relates to a moving image rate conversion device, and more particularly, to a moving image rate conversion device that converts a moving image encoded with an arbitrary transmission bit rate to a lower bit rate and outputs the converted moving image.

  The conversion of the bit rate of encoded moving image data is effective in the case where a moving image encoded at a relatively high bit rate is distributed via a network having a lower bit rate.

  Conventionally, the system shown in FIG. 11 is known as a typical bit rate conversion system for encoded moving image data. In the first prior art shown in FIG. 6A, the encoded moving image is decoded by the variable length decoding unit 51, then inverse quantized by the inverse quantization unit 52, and further by the inverse DCT transform unit 53. In this method, inverse DCT conversion is performed to return to the pixel level, and the DCT conversion unit 54, the quantization unit 55, and the variable length encoding unit 56 perform conversion again to a desired bit rate.

The second prior art shown in FIG. 2B is a conversion method disclosed in Patent Document 1, in which a coded moving image is decoded by a variable length decoding unit 61 and then requantized by a DCT. In this method, the coefficient is inversely quantized, a new quantization parameter is set, requantization is performed, and the variable length coding unit 63 recodes the coefficient.
JP 2001-186519 A

As an index for evaluating the coding efficiency of moving images, the coding cost Ji obtained by substituting the coding distortion Di (the square error between the original image and the reproduced image) and the generated code amount Ri into the following equation (1) is known. It has been. Note that λ is a constant determined based on the quantization parameter.

Ji ＝ Di ＋ λ ・ Ri (1)

  When converting the bit rate of the moving image, the moving image is returned to the pixel level in the first prior art, so that the possibility of reducing the encoding cost is increased, but re-encoding involves a large amount of calculation. Therefore, the processing time tends to be long. On the other hand, since the second prior art does not require re-encoding, the amount of calculation can be reduced and the processing time can be shortened. However, side information such as MB (macroblock) information and MV (motion vector) information is not changed. The encoding cost tends to be higher than that of the first prior art. As described above, the conventional bit rate conversion method has advantages and disadvantages, and it has been difficult to optimize encoding cost and processing time.

  The object of the present invention is to solve the above-described problems of the prior art and optimize encoding cost and processing time while minimizing visual degradation of moving images in bit rate conversion of encoded moving images. An object of the present invention is to provide a moving image rate conversion device that can perform this.

In order to achieve the above-described object, the present invention is characterized in that the following measures are taken in a moving image rate conversion device for converting the transmission bit rate of moving images.
(1) Re-encoding means for rate conversion in the process of decoding and re-encoding the moving image to the pixel level, re-quantization means for rate conversion in the process of de-quantizing and re-quantizing the moving image, It is characterized by comprising a conversion method selection means for selecting one of the re-encoding means and the re-quantization means.
(2) The conversion method selection means selects one of the conversion means based on at least one bit amount of information to be recoded and information to be requantized.
(3) The conversion method selection means compares the difference value between the quantization parameter before rate conversion and the quantization parameter after rate conversion with a reference difference value, and if the difference value exceeds the reference difference value, The encoding means is selected, and if it does not exceed, the requantization means is selected.
(4) The conversion method selection means selects the re-quantization means if the bit amount of the motion vector information (MV information) is below a predetermined first threshold, and selects the re-encoding means if it is not below It is characterized by that.
(5) If the difference value obtained by subtracting the bit amount of the prediction coding mode information (MB information) from the bit amount of the DCT coefficient exceeds the predetermined second threshold, the conversion method selection unit selects the re-quantization unit. If it does not exceed, re-encoding means is selected.
(6) The conversion method selection means determines the encoding state of the predictive coding mode information (MB information), and if MB information is properly encoded, selects the re-quantization means and encodes it appropriately. If not, the re-encoding means is selected.

According to the present invention, the following effects are achieved.
(1) As a function of converting the bit rate, rate conversion is performed in the process of decoding and re-encoding the moving image to the pixel level, re-encoding means that can reduce the encoding cost although the processing time is long, and the moving image The rate conversion is performed in the process of dequantization and requantization, and the requantization means with a short processing time is provided, although the encoding cost is inferior, and each conversion means is selectively used. Optimization with time becomes possible.
(2) Since one of the transform means is selected based on the bit amount of at least one of the information to be recoded and the information to be requantized, the information to be requantized If the requantization means is selected when the bit amount is large, the processing time can be shortened without increasing the coding cost. Further, if the re-encoding means is selected when the bit amount of information to be re-encoded is large, the encoding cost can be suppressed.
(3) If the difference value between the quantization parameter before rate conversion and the quantization parameter after rate conversion is large and the bit amount of information to be converted by the requantization means is small, the recoding means selects As a result, the encoding cost can be reduced even when the difference value is large.
(4) If the bit amount of motion vector information (MV information) is small and there is no significant difference in coding cost between the re-encoding means and the re-quantizing means, the re-quantizing means is selected. The processing time can be shortened without increasing the coding cost.
(5) Requantization when the difference between the bit amount of DCT coefficient information and the bit amount of predictive coding mode information (MB information) is large and the coding cost can be reduced even by rate conversion by requantization means Since the means is selected, a low coding cost can be achieved in a short processing time.
(6) If the coding cost of predictive coding mode information (MB information) is low and there is no significant difference in coding cost between the re-coding means and the re-quantizing means, the re-quantizing means is selected. As a result, the processing time can be shortened without increasing the coding cost.

  DESCRIPTION OF THE PREFERRED EMBODIMENTS Hereinafter, the best embodiment of the present invention will be described in detail with reference to the drawings. FIG. 1 is a functional block diagram showing the configuration of a moving image rate conversion apparatus according to the present invention. Here, a transmission rate conversion method for moving image data encoded and compressed by the MPEG method, which is an international standard for general-purpose moving image encoding, will be described, but the present invention is not limited to this, and other methods are used. The transmission rate can be converted by the same process for the encoded and compressed moving image data.

  The variable length decoding unit 1 receives a moving image in which each frame is predictively encoded in units of macroblocks in units of frames. The variable length decoding unit 1 performs variable length decoding of a moving image, and represents a quantization parameter QP, quantized DCT coefficient information COEF1, MB information representing a macroblock prediction coding mode, and a motion vector for each macroblock. MV information and various header information (sequence layer, GOP layer, picture layer, slice layer) are extracted. The target quantization parameter conversion unit 2 converts the quantization parameter QP extracted by the variable length decoding unit 1 into a quantization parameter target value (target quantization parameter) QPt after bit rate conversion.

  The re-encoding unit 4 decodes the moving image to the pixel level, and performs bit rate conversion in the process of re-encoding it. In the rate conversion based on this re-encoding method, a large amount of calculation is required for re-encoding, and there is a high possibility that the encoding cost can be kept low. The requantization unit 5 dequantizes the moving image and performs bit rate conversion in the process of requantizing the moving image. In the rate conversion based on this requantization method, processing with a small amount of calculation is possible although the coding cost increases compared to recoding.

  The conversion method selection unit 3 sends the quantization parameter QP, DCT coefficient information COEF1, MB information, MV information and various header information extracted by the variable length decoding unit 1, and the target quantization parameter conversion unit 2 The target quantization parameter QPt and the like are input, and as will be described in detail later, one of the re-encoding unit 4 and the re-quantization unit 5 is set to the information to be re-encoded and the re-quantization target. The selection is based on the amount of information bits.

  FIG. 2 is a block diagram showing an example of the requantization unit 5. The motion vector information MV is input to the difference image memory unit 501. The inverse quantization unit 502 receives the quantization parameter QP and the quantized DCT coefficient information COEF1. The quantization parameter QP and the quantized DCT coefficient information COEF1 input to the inverse quantization unit 502 are converted into DCT coefficient information COEF2 using a normal inverse quantization method. The inversely quantized DCT coefficient information COEF2 is sent to the DCT coefficient adding unit 503 and the DCT coefficient subtracting unit 507.

  DCT coefficient adding section 503 generates new DCT coefficient information COEF3 using DCT coefficient information COEF2, MB information MB, and DCT coefficient COEF4 of the difference frame input from DCT conversion section 513. The requantization unit 504 requantizes the DCT coefficient information using the DCT coefficient information COEF3 and the target quantization parameter QPt input from the DCT coefficient addition unit 503, and newly generates DCT coefficient information COEF5. . The target quantization parameter QPt is obtained by rate control in the variable length decoding unit 1, and the requantization unit 504 applies, for example, a quantization method defined by TM2 (Test Model 5) of MPEG-2. can do.

  The DCT coefficient information COEF5 generated by requantization is variable-length encoded by the variable-length encoding unit 505 together with the header information sent from the variable-length decoding unit 1 and the target quantization parameter QPt, and a new moving image is obtained. Is output as At this time, the same information as the information extracted from the original moving image is used for the header information, but information such as the bit rate, VBV buffer size, and VBV delay is recalculated according to the bit rate of the new moving image. The

  The DCT coefficient information COEF5 and the target quantization parameter QPt are input to the inverse quantization unit 506. The inverse quantization unit 506 performs inverse quantization using the DCT coefficient information COEF5 and the target quantization parameter QPt to generate DCT coefficient information COEF6. The inversely quantized DCT coefficient information COEF6 is sent to the DCT coefficient subtraction unit 507.

  The DCT coefficient subtraction unit 507 generates difference DCT coefficient information DCOEF1 by subtracting DCT coefficient information COEF6 from DCT coefficient information COEF2. The difference DCT coefficient information DCOEF1 is input to the inverse DCT conversion unit 509. In the inverse DCT conversion unit 509, information source decoding for generating a difference pixel is performed on the input DCOEF1, that is, inverse DCT conversion is performed to generate difference pixel data DPIX1. The difference pixel data DPIX1 is sent to the difference frame addition unit 510. The difference frame addition unit 510 adds the difference pixel data DPIX1 input from the inverse DCT conversion unit 509 and the difference motion compensated prediction pixel data DPIX2 input from the difference image memory unit 501 based on MB (predictive coding mode) information. As a result, new difference pixel data DPIX3 is obtained.

  In the difference image memory unit 501, if the frame coding type is an intra-picture coded picture (I picture) or a forward inter-picture predictive coded picture (P picture), each macroblock is stored at a corresponding address. . Further, in the difference image memory unit 501, when MV (motion vector) information is input from the variable length decoding unit 1, that is, when the prediction coding mode information MB of the macroblock is in the inter coding mode, the difference image memory unit 501 Using the difference pixel data DPIX3 and the motion vector MV, difference motion compensated prediction pixel data DPIX2 is obtained by the same method as the normal motion compensation prediction, and this is sent to the difference frame addition unit 510. If the MB information is the intra coding mode, no motion vector is present, so motion compensation prediction is not performed. In the DCT conversion unit 513, the difference motion compensated prediction pixel data DPIX2 is information source encoded, that is, DCT converted, and DCT coefficient information COEF4 of the difference frame is generated. The DCT coefficient information COEF4 of the difference frame is input to the DCT coefficient adding unit 503.

  According to the re-quantization scheme having such a configuration, the DCT coefficient subtraction unit 507 subtracts the converted DCT coefficient information COEF6 from the DCT coefficient information COEF2 before the transmission rate conversion, thereby obtaining an error before and after the transmission rate conversion. Difference DCT coefficient information DCOEF1 is generated, and this difference is subjected to inverse DCT conversion by the inverse DCT conversion unit 9, and an error amount of the reference image before and after transmission rate conversion is obtained by motion compensation prediction. Is subjected to DCT conversion by the DCT conversion unit 13 and fed back to the DCT coefficient of the prediction error image before the transmission rate conversion, so that the error can be corrected and accumulation of errors due to re-quantization can be prevented. Become.

  FIG. 3 is a block diagram showing an example of the re-encoding unit 4. The inverse quantization unit 401 receives a quantization parameter QP, quantized DCT coefficient information COEF1, predictive coding mode information MB, and header information. These pieces of information pass through an inverse quantization unit 401, an inverse DCT conversion unit 402, and an adder 412 and are converted into video information decoded to a pixel level. This video information is input to the image rearrangement unit 403 and also input to the image memory unit 408. The frame data read from the image memory unit 408 is motion compensated by the motion compensation prediction unit 407 using the motion vector information MV, and then input to the adder 412.

  The DCT conversion unit 404 converts the input from the subtracter 413 into DCT coefficients in units of blocks. The DCT coefficient thus generated is input to the quantization unit 405. In the quantization unit 405, requantization for bit reduction is performed based on the target quantization parameter QPt. The DCT coefficient re-quantized by the quantization unit 405 is variable-length encoded by the variable-length encoding unit 406 and is output as a moving image after bit rate conversion.

  The output of the quantization unit 405 is input to the adder 414 through the inverse quantization unit 409 and the inverse DCT conversion unit 410. The adder 414 generates a prediction screen, and the generated prediction screen is input to the image memory unit predictor 411. The prediction screen read from the image memory unit predictor 411 is input to the subtracter 413 and the adder 414.

  FIG. 4 is a flowchart showing the selection procedure of the bit rate conversion method in the conversion method selection unit 3. In step S1, the current quantization parameter QP is subtracted from the target quantization parameter QPt to obtain the difference value ΔQP. . In step S2, the difference value ΔQP obtained in step S1 is compared with a predetermined reference difference value ΔQPref. If the difference value ΔQP exceeds the reference difference value ΔQPref, the process proceeds to step S3 and a re-encoding method is selected. On the other hand, if the difference value ΔQP does not exceed the reference difference value ΔQPref, the process proceeds to step S4, and the requantization method is selected.

  5 and 6 are diagrams schematically showing a mechanism for optimizing the coding cost and the processing speed by selecting the bit rate conversion method based on the quantization parameter QP as described above.

  That is, as shown in FIG. 5A, the bit amount generated by encoding depends on the bit amount a of DCT coefficient information and the bit amount b of side information including MB information and MV information. In the re-quantization by the re-quantization unit 5, only the DCT coefficient information is reduced and the side information is fixed without being changed, whereas in the re-encoding by the re-encoding unit 4, The bit amount is optimized including not only the DCT coefficient information but also the side information.

  In the requantization, the quantization step increases as the quantization parameter (QP value) increases, so that although the quantization error increases, the coding efficiency is improved. If the bit amount a of the DCT coefficient information decreases as the coding efficiency improves, the ratio of the side information to the generated bit amount increases as shown in FIG. Therefore, when the QP value increases to some extent, the degree of decrease in the amount of generated bits with respect to the increase rate of the QP value becomes small, so that the effect of improving the coding efficiency decreases.

  On the other hand, in the re-encoding method, as shown in (c) of the figure, since encoding is performed so that the generated bit amount is minimized including not only DCT coefficient information but also side information, Higher encoding efficiency can always be expected than the requantization method. Such a tendency is conspicuous particularly in an encoder that optimizes a macroblock prediction mode based on “rate-distortion optimization” that has been adopted by encoders such as H.264 / AVC.

  FIG. 6 shows a case where the target QP value (QPt) is changed from “15” to “36” with the H.264 / AVC content encoded with the quantization parameter (QP value) “12” as input material. The amount of generated bits is shown for each rate conversion method, the solid line indicates the re-quantization method, and the wavy line indicates the re-encoding method.

  In the illustrated example, the amount of bits generated by the re-encoding method is lower than that of the re-quantization method in the range where the QP value after conversion is around “27” and the QPt value is larger than that. Therefore, the encoding difference (encoding efficiency) is to set the reference difference value ΔQPref to “15 (27-12)” and select a re-encoding method if the difference value ΔQP exceeds “15”. Desirable from a viewpoint. However, the re-quantization method has a smaller calculation amount than the re-encoding method, and the processing time can be reduced to about half. Therefore, if the difference value ΔQP does not exceed “15”, there is no difference in encoding cost, and it is desirable from the viewpoint of processing speed to employ the requantization method.

  As described above, in the present invention, when comparing the encoding costs of the re-encoding method and the re-quantization method, it is newly found that the superiority depends on the magnitude of the QP value, and this encoding cost and processing The bit rate conversion method is dynamically selected based on the relationship with the speed.

  In the above-described first embodiment, the bit rate conversion method is selected based on the quantization parameter (QP value). However, the present invention is not limited to this, and other conditions may be satisfied. Based on this, a bit rate conversion method may be selected.

  FIG. 7 is a flowchart showing a second embodiment of the selection procedure in the conversion method selection unit 3. In this embodiment, the conversion method paying attention to MV information (information on motion vectors) which is one of side information. There is a feature in that it is selected.

  In step S21, the bit amount of the MV information is compared with a predetermined threshold value. If the bit amount of the MV information is below the threshold value, the process proceeds to step S23 and the requantization method is selected. On the other hand, if the bit amount of the MV information exceeds the threshold, the process proceeds to step S22 and the re-encoding method is selected.

  That is, if the bit amount of MV information is small, the bit amount of side information will be small, so there is a big difference in coding cost between re-encoding method that optimizes including side information and re-quantization method that does not change side information. Does not occur. Therefore, when the bit amount of MV information is less than a predetermined reference value, it is desirable to adopt a requantization method from the viewpoint of processing speed. The present embodiment is effective in the case of an intra-coded image or MB, or at the time of rate conversion of a moving image having no MV information using the direct mode in MB.

  FIG. 8 is a flowchart showing a third embodiment of the selection procedure in the conversion method selection unit 3. In this embodiment, the conversion method is selected by paying attention to MB information which is one of side information. There is a feature in the point.

  In step S31, whether or not MB information is optimally encoded is verified based on the H.264 / AVC rate-distortion optimization method. If the MB information is optimally encoded and the encoding cost indicates the minimum value, the process proceeds to step S33 and the requantization method is selected. On the other hand, if the MB information is not optimally encoded, the process proceeds to step S32, and a re-encoding method is selected.

  In other words, if MB information is already optimally encoded, the MB information encoding cost cannot be further increased. Therefore, even in the re-encoding method that optimizes including side information, requantization that does not change side information is performed. Even in the coding scheme, there is no significant difference in coding efficiency. Therefore, when MB information is optimally encoded, it is desirable to adopt a requantization method from the viewpoint of processing speed. The present embodiment is effective when the motion vector is (0, 0) and the block size is the maximum, or when the rate conversion is performed on a moving image whose MB type is a skipped macroblock (copy of the previous image).

  FIG. 9 is a flowchart showing a fourth embodiment of the selection procedure in the conversion method selection unit 3, and this embodiment is characterized in that the conversion method is selected by paying attention to the bit amount of the DCT coefficient. is there.

  In step S41, a difference value obtained by subtracting the bit amount of the MB information from the bit amount of the DCT coefficient is compared with a predetermined threshold value. If the difference value exceeds the threshold value, the process proceeds to step S43 and the requantization method is selected. On the other hand, if the difference value does not exceed the threshold value, the process proceeds to step S42, and the re-encoding method is selected.

  In other words, even if the bit amount of DCT coefficient is large and MB information is optimized by re-encoding (and the bit amount of DCT coefficient increases accordingly), the ratio of DCT coefficient information to the total bit amount If it is high, it is not necessary to adopt a re-encoding method that optimizes including side information, and re-quantization that reduces only the bit amount of DCT coefficient information can also obtain sufficient encoding efficiency, so re-quantization It is desirable to adopt a method.

  In the above embodiment, the bit rate conversion method is selected based on one parameter. However, the present invention is not limited to this, and the first example shown in FIG. As in the fifth embodiment, the selection may be made by appropriately combining the conditions of the above-described embodiments.

  In the present embodiment, in step S51 (S21), the bit amount of the MV information is compared with a predetermined threshold value. If the bit amount of the MV information is below the threshold value, the process proceeds to step S57 (S23) and the requantization method is selected. If the bit amount of the MV information exceeds the threshold value, the process proceeds to step S52.

  In step S52 (S31), it is determined whether the MB information is optimally encoded. If the MB information is optimally encoded, the process proceeds to step S57 and a requantization method is selected. If the MB information is not optimally encoded, the process proceeds to step S53.

  In step S53 (S41), a difference value obtained by subtracting the bit amount of the MB information from the bit amount of the DCT coefficient is compared with a predetermined threshold value. If the difference value exceeds the threshold value, the process proceeds to step S57 and the requantization method is selected. If the difference value does not exceed the threshold value, the process proceeds to step S54.

  In step S54 (S1), a difference value ΔQP is obtained by subtracting the current quantization parameter QP from the target quantization parameter QPt. In step S55 (S2), the difference value ΔQP obtained in step S54 is compared with a predetermined reference difference value ΔQPref. If the difference value ΔQP exceeds the reference difference value ΔQPref, the process proceeds to step S56 and a re-encoding method is selected. On the other hand, if the difference value ΔQP does not exceed the reference difference value ΔQPref, the process proceeds to step S57 and the requantization method is selected.

It is the block diagram which showed the structure of the principal part of the moving image rate conversion apparatus concerning this invention. It is the block diagram which showed an example of the requantization part of FIG. It is the block diagram which showed an example of the re-encoding part of FIG. It is the flowchart which showed the selection procedure of 1st Embodiment of the rate conversion system in the conversion system selection part of FIG. It is the figure which showed typically the structure in which encoding efficiency and a processing speed are optimized by selecting a rate conversion system based on the quantization parameter QP. FIG. 5 is a diagram illustrating a relationship between a target quantization parameter (QPt) and a generated bit amount for each rate conversion method. It is the flowchart which showed the selection procedure of 2nd Embodiment of the rate conversion system in the conversion system selection part of FIG. It is the flowchart which showed the selection procedure of 3rd Embodiment of the rate conversion system in the conversion system selection part of FIG. It is the flowchart which showed the selection procedure of 4th Embodiment of the rate conversion system in the conversion system selection part of FIG. It is the flowchart which showed the selection procedure of 5th Embodiment of the rate conversion system in the conversion system selection part of FIG. It is a block diagram of a prior art.

Explanation of symbols

  DESCRIPTION OF SYMBOLS 1 ... Variable length decoding part, 2 ... Target quantization parameter conversion part, 3 ... Conversion system selection part, 4 ... Re-encoding part, 5 ... Re-quantization part,

Claims (6)

In a moving image rate conversion device that converts a transmission bit rate of moving images,
Re-encoding means for performing rate conversion in the process of decoding and re-encoding the moving image to a pixel level;
Requantization means for rate conversion in the process of dequantizing and requantizing the moving image;
A rate conversion apparatus for moving images, comprising: a conversion method selection unit that selects one of the re-encoding unit and the re-quantization unit.   The conversion method selection unit selects one of the conversion units based on at least one bit amount of information to be re-encoded and information to be re-quantized. The moving image rate conversion apparatus described.   The conversion method selection means compares the difference value between the quantization parameter before conversion and the quantization parameter after conversion with a reference difference value, and selects the re-encoding means if the difference value exceeds the reference difference value. 3. The moving image rate conversion apparatus according to claim 2, wherein if the difference value does not exceed a reference difference value, a re-quantization unit is selected.   The conversion method selection means selects the re-quantization means if the bit amount of motion vector information (MV information) is below a predetermined first threshold value, and the bit amount of MV information must be below the first threshold value. 3. The moving image rate conversion apparatus according to claim 2, wherein re-encoding means is selected.   The conversion method selection means selects the requantization means if the difference value obtained by subtracting the bit amount of the predictive coding mode information (MB information) from the bit amount of the DCT coefficient exceeds a predetermined second threshold, and DCT 3. The moving image rate conversion apparatus according to claim 2, wherein re-encoding means is selected if the bit amount of the coefficient does not exceed the second threshold value.   The conversion method selection means determines the encoding state of the predictive coding mode information (MB information), selects the re-quantization means if the MB information is properly encoded, and appropriately encodes the MB information. 3. The moving image rate conversion apparatus according to claim 2, wherein re-encoding means is selected if not.

Images