JP4072859B2 - Video information re-encoding device - Google Patents

Description

  The present invention relates to a video information re-encoding apparatus, and in particular, video information capable of suppressing image quality deterioration in a portion where image quality deterioration is conspicuous, such as an edge portion or a flat portion, when converting the bit rate of encoded video information. The present invention relates to a re-encoding device.

  Currently, there are various network networks such as broadband and mobile in networks that transmit video information. In addition, there are various terminals such as a personal computer, a mobile, and a PDA for reproducing the video information. In such various network environments and playback environments, in order to share video content and use it efficiently, it is necessary to convert video information into coding parameters suitable for the network environment and playback environment. is necessary. Bit rate conversion of video information is one type of encoding parameter conversion.

  The most primitive method for bit rate conversion of encoded video information is called simple re-encoding. In simple re-encoding, encoded video information is decoded and the decoded video information is encoded at different bit rates. When video information is bit-rate converted by simple re-encoding, a quantization error is increased and image quality is deteriorated.

  Patent Document 1 and Non-Patent Document 1 propose a re-encoding device that suppresses image quality degradation associated with re-encoding. FIG. 5 is a block diagram showing this re-encoding device. In FIG. 5, an entropy decoding unit 101, an inverse quantization unit 102, an inverse DCT unit 103, an adder 104, and a frame memory 105 constitute a decoding unit that decodes encoded video information (input stream).

  Also, the subtractor 106, the DCT unit 107, the quantization unit 108, the inverse quantization unit 109, the inverse DCT unit 110, the adder 111, the frame memory 112, the rate control unit 113, and the entropy encoding unit 114 are re-encoded. Parts.

  The encoded video information input from the input terminal (INPUT) 100 is first input to the decoding unit and decoded to the pixel level. The entropy decoding unit 101 of the decoding unit also outputs a quantization parameter (QP).

  The re-encoding unit converts the frame data from the decoding unit into DCT coefficients again in units of blocks, further generates quantized and variable-length encoded video information that has been bit-rate converted, and outputs an output terminal (OUTPUT) Output from 115.

When the quantization parameter after the bit rate conversion falls between 1 to 2 times the quantization parameter (QP) before the bit rate conversion output from the entropy encoding unit 114, the rate control unit 113 performs the quantization. Forcibly shift parameter by 1 or 2 times. The quantization unit 114 prevents image quality degradation by quantizing the DCT coefficient from the DCT unit 107 with the shifted quantization parameter (QP ′).
JP 2001-204028 A "Requantization with Zone Processing for MPEG Recoding" IEICE Tech.Vol.J83-D-II No.9 pp.1879-1887 2000/9

  According to the re-encoding device proposed in Patent Document 1 and Non-Patent Document 1, it is possible to suppress image quality deterioration associated with bit rate conversion. However, the image quality of the image has a feature that deterioration is conspicuous if there is even a slight noise mixture or level change at the edge or flat part in the screen. In the proposed re-encoding apparatus, The part where the image quality degradation is visually conspicuous is not taken into consideration, and no countermeasure is taken for it.

  The object of the present invention is to re-use the motion vector, coding mode information, and quantization parameter when converting the bit rate of the encoded video information, and to improve the image quality in a portion where image quality degradation is conspicuous, such as an edge portion or a flat portion. An object of the present invention is to provide a video information re-encoding device capable of suppressing deterioration.

In order to solve the above problems, the present invention provides:
In a video information re-encoding device for converting the bit rate of encoded video information, first means for reusing motion vector information and intra-screen / inter-screen encoding mode information, and DCT coefficient re-quantization parameters are input. A second means for prohibiting the re-quantization parameter from being more than one time and less than twice the quantization parameter of the input stream, with a region that is more than one time and less than twice the quantization parameter of the stream as a prohibited region; When the quantization scaling factor r is given, the ratio p: 1-p (0 ≦ p ≦ 1) of the number of macroblocks to be multiplied by 1 and the number of macroblocks to be multiplied by 2 is given by (1) According to the third means calculated by the equation and the macroblock belonging to the prohibited area, the quantization parameter of the input stream is changed to a macroblock having a high edge strength according to the ratio calculated by the third means. There is a first feature in that a fourth means for preferentially assigning a requantization parameter that is one time larger than that of the data is provided.
p = 2 − r (1 <r <2) (1)

Also, in the present invention, the fourth means may be configured to input a stream into a macroblock with high flatness instead of a macroblock with high edge strength, or into both a macroblock with high edge strength and a macroblock with high flatness. A second feature is that a re-quantization parameter that is 1 times the quantization parameter is preferentially assigned.

  Further, the present invention obtains the initial value of the requantization parameter by multiplying the quantization parameter of the input frame by a quantization scaling coefficient calculated by (bit rate before conversion) / (bit rate after conversion). There is a third feature in that means are provided.

The present invention further comprises means for detecting edge strength based on a calculated value of a Sobel Operator using eight blocks adjacent in the vertical, horizontal, and diagonal directions, and the fourth means includes an edge The re-quantization parameter that is one time the quantization parameter of the input stream is preferentially assigned to a high-strength macroblock or to both a macroblock with high edge strength and a macroblock with high flatness . There are four features.

The present invention further comprises means for detecting flatness using the sum of all components of the DCT coefficients in the macroblock , and the fourth means is a macroblock with high flatness or a macro with high edge strength. A fifth feature is that a requantization parameter that is one time the quantization parameter of the input stream is preferentially assigned to both the block and the macroblock with high flatness .

Furthermore, in the present invention, the fourth means preferentially assigns a requantization parameter that is one time the quantization parameter of the input stream to both a macroblock with high edge strength and a macroblock with high flatness. , and the upon the assignment, there is a sixth feature of the that <br/> point prioritize according to the value of the weighting function of the edge intensity and flatness as shown in the following equation (2).
Priority = α × (edge strength) + (1-α) × flatness (0 ≦ α ≦ 1) (2)

  According to the present invention, when converting the bit rate of encoded video information, the image quality deterioration of a portion where image quality deterioration is conspicuous, such as an edge portion or a flat portion, while reusing motion vectors, encoding mode information, and quantization parameters. Can be suppressed.

  As a result, video distribution to various network environments and playback environments can be performed with less degradation in image quality and accurately in accordance with the target rate.

  The present invention will be described below with reference to the drawings. FIG. 1 is a block diagram showing an embodiment of a video information re-encoding device according to the present invention. The same reference numerals are given to the same or equivalent parts as FIG.

  In FIG. 1, an entropy decoding unit 101, an inverse quantization unit 102, an inverse DCT unit 103, an adder 104, a frame memory 105, a motion compensation unit 116, and an edge detection unit 117 decode encoded video information (input stream). A decoding unit is configured.

  Also, the subtractor 106, DCT unit 107, quantization unit 108, inverse quantization unit 109, inverse DCT unit 110, adder 111, frame memory 112, motion compensation unit 118, rate control unit 113, and entropy encoding unit 114 Constitutes a re-encoding unit that outputs encoded video information subjected to bit rate conversion.

  First, the operation of the decoding unit will be described. The encoded video information input from the input terminal (INPUT) 100 is input to the decoding unit. The entropy decoding unit 101 of the decoding unit performs variable length decoding on the input encoded video information. The entropy decoding unit 101 outputs a quantization parameter (QP), buffer transition information (BT), motion vector information (MV), and coding mode information (not shown) in addition to the video information.

  The video information variable-length decoded by the entropy decoding unit 101 is passed through the inverse quantization unit 102, the inverse DCT unit 103, and the adder 104, so that the video information decoded to the pixel level is output. This video information is input to the encoding unit and also input to the frame memory 105.

  The frame data read from the frame memory 105 is motion compensated by the motion compensation unit 115 using the motion vector information (MV), and then input to the adder 104. The edge detection unit 117 detects the edge part in the screen using the data of the DCT coefficient region inversely quantized by the inverse quantization unit 102.

  Next, the operation of the re-encoding unit will be described. The frame data output from the decoding unit is input to the re-encoding unit and re-encoded at a different bit rate. In the re-encoding here, the motion vector information (MV) and the encoding mode information obtained by the decoding unit are reused as they are.

  The subtracter 106 of the re-encoding unit transmits the difference between the input frame data and the predicted frame data in the case of an inter-frame encoded frame, and the input frame data as it is in the case of an intra-frame encoded frame.

  The DCT unit 107 converts the input from the subtractor 106 into DCT coefficients in units of blocks. The DCT coefficient thus generated is input to the quantization unit 108. The quantization unit 108 performs re-quantization for bit reduction under the control of the rate control unit 113. Details of the rate control by the rate control unit 113 will be described later. The DCT coefficients re-quantized by the quantization unit 108 are variable-length encoded by the entropy encoding unit 114, and transmitted from the output terminal (OUTPUT) 115 as encoded video information subjected to bit rate conversion.

  The output of the quantization unit 108 is input to the adder 111 through the inverse quantization unit 109 and the inverse DCT unit 110. The adder 111 generates a prediction screen, and the generated prediction screen is input to the frame memory 112. Frame data as a prediction screen read from the frame memory 112 is input to the adder 111 and motion-compensated by the motion compensation unit 118, and then input to the subtractor 107.

  Here, since it is not assumed that the image size or the frame rate is changed, the motion vector information (MV) output from the entropy decoding unit 101 can be reused as it is for motion compensation in the motion compensation unit 118. .

  FIG. 2 is a flowchart showing an example of rate control by the rate control unit 113. First, when the encoded video information is decoded by the decoding unit, the quantization parameter (QP), buffer transition information (BT), motion vector information (MV), and inverse quantization unit 102 output from the entropy decoding unit 101 Holds the DCT coefficient output from (S1).

Next, an initial value r ₀ of the quantization scaling factor r for each frame is calculated using the quantization parameter (QP). Thereafter, the quantization scaling coefficient r is updated in units of frames using the quantization parameter (QP) and the buffer transition information (BT) (S2). The determination of the quantization scaling factor r has been described in detail in the prior application (Japanese Patent Application No. 2003-206482). The outline of the determination is as follows.

First, the initial value r ₀ of the quantization scaling coefficient r is calculated by the following equation (3).
r ₀ = (bit rate before conversion) / (bit rate after conversion) (3)
For example, if the bit rate of the input encoded video information is 8 Mbps and the bit rate of the output encoded video information is 5 Mbps, the initial value r ₀ of the quantization scaling factor r is 1.6.

The quantization scaling factor r is updated in units of frames according to the following equation (4). Here, p ₂ is a parameter of the variation amount of the quantization scaling factor r, determined by evaluation based on the average value and the standard deviation of {(bhr '(tn) -blr (tn)) / blr (tn)} Is done. As an example, p ₂ = 0.1 can be set. Equation (4) below updates the quantization scaling factor r in the VBV (Video Buffering Verifier) buffer so that the buffer occupancy blr (tn) of the output encoded video information becomes the ideal buffer occupancy bhr ′ (tn). It means to do. Here, tn is the decoding time of the nth image.
r _{n + 1} = r _n × [1 + {(bhr ′ (tn) −blr (tn)) / blr (tn)} × p ₂ ] (n ≧ 0) (4)

  Next, an edge portion of the screen is detected using the DCT coefficient (S3). The edge portion can be detected using the characteristics of the DCT coefficient. For example, since the DCT coefficient of the block including the edge has a feature that the coefficient is concentrated on the first and second rows and columns, the edge is calculated by calculating the sum of the DCT coefficients of the first and second rows and columns. It can be detected. Further, since there is a tendency to include an edge in a region where the prediction error amount is large, the edge can also be detected by calculating the sum of DCT coefficients. Furthermore, an edge can also be detected using a Sobel operator.

  When using Sobel Operator, for example, DC component of DCT coefficients in the surrounding nine blocks including own block is extracted, and gradient magnitude is calculated using Sobel Operator with the extracted nine components as elements. Good.

  The average value of the values calculated in this way in the macroblock or the value of one specific block included in the macroblock is set as the edge strength of the macroblock. As described above, the processing load can be reduced by detecting edges in the DCT coefficient region instead of the pixel region. When the edge is detected, labels are attached in order from the edge having the highest strength.

Next, it is determined whether or not the quantization scaling factor r is 2>r> 1 (S4). If it is determined in S4 that the quantization scaling factor r is not 2>r> 1, the requantization parameter QP ′ is calculated by the following equation (5), and requantization is performed using the requantization parameter QP ′. (S5). That is, the quantization scaling factor r in units of frames calculated in S2 is used as it is.
QP ′ = QP × r (5)

If it is determined in S4 that the quantization scaling factor r is 2>r> 1, the number of macroblocks for which the requantization parameter is set to r = 1 and the macroblock for which r = 2 are set for the macroblock of the frame. The ratio of the numbers p: 1-p is calculated by the following equation (6) (S6).
p = 2-r (1 <r <2) (6)

  For example, when the quantization scaling factor r is 1.6, the ratio p: 1-p is 4: 6, and the macroblock 6 that sets r = 2 to the macroblock 4 that sets the requantization parameter to r = 1. It becomes the ratio. The ratio p calculated here dynamically changes in units of frames according to the output bit rate target buffer transition and the quantization scaling factor r.

Subsequently, the quantization scaling factor 1 is given in order from the macroblock with the large edge strength detected in S2 until the ratio becomes p, and the quantization scaling factor 2 is given to the macroblocks after reaching that ratio. . Assuming that the quantization scaling coefficient given for each macroblock in this way is r ′, the rate control unit 113 calculates the requantization parameter QP ′ by the following equation (7), and the quantization unit 108 Using this requantization parameter QP ′, the frame is requantized (S7). Re-quantization with the quantization scaling factor r ′ = 1 means that re-quantization is performed with the same quantization parameter as the input encoded video information.
QP ′ = QP × r ′ (7)

  After requantizing the frame in S5 and S7, it is determined whether or not the input video is finished (S8). If it is determined that the input video has been completed, the process is terminated. If it is determined that there are still frames of the input video, it is updated to the next frame (S9) and the steps from S1 are repeated.

  As described above, since the same requantization parameter as the quantization parameter of the input encoded video information is preferentially given to the macroblock including the edge at the time of requantization, data loss at the edge portion is prevented. Reduced. Also, since motion vectors are shared before and after bit rate conversion, motion detection processing is reduced, and the processing burden associated with overall encoding is reduced.

  Although the reduction of data loss at the edge portion of the screen has been described above, image quality deterioration is also noticeable due to a slight difference in flatness (flatness) in the flat portion of the screen. The same requantization parameter as the quantization parameter of the input encoded video information may be preferentially given to both of the flat portions. The flatness of the screen can be detected by using the DCT coefficient similarly to the edge strength, for example, by calculating the sum of all the components of the DCT coefficient in the macroblock.

Also, as shown in the following equation (8), priorities are given according to the values of the weighting function of edge strength and flatness, and if α is made variable here, the edge portion is given priority or the flat portion is given priority. Or can be set arbitrarily.
Priority = α × (edge strength) + (1-α) × flatness (0 ≦ α ≦ 1) (8)

  In addition, a method described in the above-mentioned prior application (Japanese Patent Application No. 2003-206482) is combined with a method of tracking the buffer transition of the output encoded video information (output stream) with the buffer transition of the input encoded video information (input stream). Thus, the target bit rate can be achieved by satisfying the rate followability and the limitation of the VBV buffer occupation amount.

  Although the embodiments have been described above, the present invention is not limited to the above-described embodiments, and various modifications are possible, and modifications that do not depart from the gist of the present invention are also included in the present invention.

  For example, the re-encoding function in the above embodiment can be configured as a program. In this case, as shown in FIG. 3, an MPEG video re-encoding program is stored in a computer-readable storage medium 300, and this MPEG video re-encoding program is read by the computer 200 and executed. can do. Here, the MPEG video re-encoding program stored in the storage medium 300 is a program for realizing the MPEG video re-encoding function shown in FIG.

  A computer 200 shown in FIG. 3 includes a CPU 201 that executes the MPEG video re-encoding program, an input device 202 such as a keyboard and a mouse, a ROM (Read Only Memory) 203 that stores various data, an operation parameter, and the like. It consists of a RAM (Random Access Memory) 204 for storing, a reading device 205 for reading an MPEG video re-encoding program from the storage medium 300, an output device 206 such as a display and a printer, and a bus BUS connecting each part of the device. ing.

  The CPU 201 reads the MPEG video re-encoding program stored in the storage medium 300 via the reading device 205, and then executes the re-encoding program to perform the above-described MPEG video re-encoding process. Do. The recording medium 300 includes a portable storage medium such as an optical disk, a flexible disk, and a hard disk, and also includes a transmission medium that temporarily stores data such as a network.

  FIG. 4 shows the result of an experiment conducted to verify the difference between the present invention and the effect of simple re-encoding. FIG. 4A shows the conditions of the input encoded video information, and the bit rate is 8 Mbps. FIG. 4B shows the conditions of the output encoded video information, and the bit rates are two types of 6 Mbps and 4 Mbps. FIG.4 (c) shows an experimental result by average PSNR. It can be seen that the average PSNR is improved by preferentially assigning bits to the edge portion and the flat portion, and the image quality deterioration of the edge portion or the flat portion such as the image quality deterioration is suppressed.

  The present invention is widely applicable to a one-source / multi-use content authoring system for BB / mobile, a BB / mobile seamless transcoder, an MPEG video conversion system, a broadband multipoint video distribution system, and the like.

It is a block diagram which shows embodiment of the video information re-encoding apparatus based on this invention. It is a flowchart which shows the example of the rate control by a rate control part. It is a block diagram of a system configuration when the present invention is configured as a video information re-encoding program. It is a figure which shows the experimental result performed in order to demonstrate the difference of this invention and the effect of simple re-encoding. It is a block diagram which shows the conventional re-encoding apparatus.

Explanation of symbols

DESCRIPTION OF SYMBOLS 100 ... Input terminal, 101 ... Entropy decoding part, 102 ... Inverse quantization part, 103 ... Inverse DCT part, 104 ... Adder, 105, 112 ... Frame memory, ..Subtracter, 107 ... DCT section, 108 ... Quantization section, 109 ... Inverse quantization section, 110 ... Inverse DCT section, 111 ... Adder, 113 ... Rate control 114, entropy coding unit, 115 ... output end, 116 ... motion compensation unit, 117 ... edge detection unit, 118 ... motion compensation unit, 200 ... computer, 201 ..CPU, 202 ... input device, 203 ... ROM, 204 ... RAM, 205 ... reading device, 206 ... output device 206, 300 ... storage medium

Claims (6)

In a video information re-encoding device for converting the bit rate of encoded video information,
First means for reusing motion vector information and intra-screen / inter-screen coding mode information;
The area where the DCT coefficient requantization parameter is more than 1 time and less than 2 times the quantization parameter of the input stream is set as a prohibited area, and the requantization parameter is more than 1 time and less than 2 times the quantization parameter of the input stream. A second means for prohibiting
When the quantization scaling factor r is given, the ratio p: 1-p (0 ≦ p ≦ 1) of the number of macroblocks to be multiplied by 1 and the number of macroblocks to be multiplied by 2 is given by (1) A third means for calculating by an equation;
For a macroblock belonging to the prohibited area, the following third ratio calculated by means of, high macroblocks edge strength, a fourth assigning 1x requantization parameters of the quantization parameter of the input stream to preferentially A video information re-encoding device comprising: means.
p = 2 − r (1 <r <2) (1) In the fourth means, instead of a macroblock having a high edge strength, a macroblock having a high flatness, or both a macroblock having a high edge strength and a macroblock having a high flatness, one of the quantization parameters of the input stream is set. 2. The video information re-encoding device according to claim 1, wherein a double re-quantization parameter is preferentially assigned. Means for obtaining an initial value of a requantization parameter by multiplying a quantization parameter of an input frame by a quantization scaling coefficient calculated by (bit rate before conversion) / (bit rate after conversion). The video information re-encoding device according to claim 1 or 2, characterized by the above-mentioned. The fourth means is a macroblock having high edge strength, comprising means for detecting edge strength based on a calculated value of a Sobel operator using eight blocks adjacent vertically, horizontally, and diagonally. , or to both high macroblocks and flatness of the edge strength is high macroblock claims 1, characterized in that the re-quantization parameter 1 times the quantization parameter of the input stream is to allocate preferentially 4. The video information re-encoding device according to any one of 3) . Means for detecting flatness using the sum of all components of DCT coefficients in the macroblock , wherein the fourth means is a macroblock having a high flatness, or a macroblock having a high edge strength and a high flatness. both the macroblock, the video information re-encoding apparatus according to claim 2 or 3, characterized in that assigning preferentially requantization parameters 1 times the quantization parameter of the input stream. The fourth means preferentially assigns a requantization parameter that is one times the quantization parameter of the input stream to both a macroblock with a high edge strength and a macroblock with a high flatness. 4. The video information re-encoding device according to claim 2, wherein priorities are assigned according to values of a weighting function of edge strength and flatness expressed by the following equation (2).
Priority = α × (edge strength) + (1-α) × flatness (0 ≦ α ≦ 1) (2)

Images