CN106454343B - The H.265 fast transcoding method of mapping is divided based on code stream bit number and block - Google Patents

Abstract

The present invention provides a kind of H.265 fast transcoding method that mapping is divided based on code stream bit number and block, the information in the code stream of input is made full use of to instruct H.265 to encode, reduce the operand of block division and model selection, largely to shorten the transcoding time, while also guaranteeing the subjective and objective visual quality of transcoding rear video.H.264/AVC or H.263++ code stream format of the present invention is not limited to H.265 isomorphism transcoding, and technology will support that, MPGE even earlier is respectively for the isomery transcoding of standard.

Description

H.265 fast transcoding method based on bit number of code stream and block division mapping

technical field

The present invention relates to video image coding technology, in particular to H.265 fast video transcoding technology.

Background technique

For the transcoding application of the new generation video codec standard H.265, the code stream is first decoded into YUV, Y4M and other formats as the source input, and then encoded with the H.265 encoder, which is called the full decoding and full encoding method. .

There are two types of H.265 video transcoding: one is homogeneous video stream transcoding, and the other is heterogeneous video stream transcoding. Homogeneous video transcoding is the conversion between code streams compressed according to the same video coding standard, while heterogeneous transcoding is the conversion between code streams compressed according to different video coding standards.

Block coding is used in H.265. Different from previous generations of encoders, the H.265 coding standard adopts a more flexible coding tree unit CTU (Coding Tree Unit) based on an adaptive quad-tree structure. The CTU can be decomposed into several square coding unit CUs (Coding Units) according to the structure of the quadtree, and the four CUs in the same layer must be of the same size. The size of the CU can be 8×8, 16×16, 32×32, 64×64. If the CTU is not decomposed, the CTU contains only one CU. Therefore, a CTU has at most 4 layers and at least 1 layer. H.265 stipulates that the size of the topmost CU in the CTU is 64×64, and the topmost CU is also called the largest coding unit (LCU). Without loss of generality, a quadtree is used to describe the division of square blocks.

CTU has three properties:

(1) Any node in the CTU has and only has 4 or 0 child nodes;

(2) Any node has one and only one parent node, and its parent node can be found according to the child node (except the root node of the 0th layer);

(3) Knowing one of the four sibling nodes of the same parent node must be able to deduce the remaining sibling nodes.

The full decoding and full encoding method is the easiest to implement and can maintain good performance in terms of rate distortion, bit rate, and subjective quality. However, the implementation process of the full decoding and full encoding method is extremely time-consuming. Especially in the transcoding process, due to unknown block division, it is necessary to traverse various situations or select a large number of block situations, which is often not desirable in actual production applications. Therefore, the fast video transcoding method has a very strong practical urgency.

In addition, when collecting and compressing video in different periods and different devices, the video coding formats used are different, resulting in various code stream formats input by the source.

SUMMARY OF THE INVENTION

The technical problem to be solved by the present invention is to provide a method that can make full use of the information in the input code stream to guide H.265 encoding to achieve fast H.265 transcoding.

The technical solution adopted by the present invention to solve the above technical problems is that the H.265 fast transcoding method based on the number of code stream bits and block division mapping includes the following steps:

The first step is to select a decoder to decode the source code stream according to the information of the input source code stream, generate a YUV sequence correspondingly according to the decoded information, and count the number of bits Bit, quantization parameter QP and encoding type of each coding unit CU in the source code stream;

The second step is to judge whether the encoding type of the source code stream is H.265, if so, go directly to the third step, if not, after realizing the bit number mapping to H.265 according to Bit and QP, go to step three;

R ₂₆₅ | _QP = a ₁ · R _other ² | _QP + a ₂ · R _other | _QP + a ₃

Among them, R ₂₆₅ | _QP represents the number of bits of a CU in the H.265 code stream under the current QP, and R _other | _QP represents the bits of the corresponding CU in the encoded code stream using a coding standard other than H.265 under the specified QP Number, a ₁ , a ₂ , a ₃ are three empirical parameters;

In the third step, the 0th frame of the YUV sequence generated by decoding is encoded in the I mode of the H.265 encoder, the first frame is encoded in the P mode of the H.265 encoder, and the remaining frames are first predicted and divided into blocks and then H.265 encoding is performed;

Among them, the steps of predicting the block are as follows:

1) Obtain the number of bits of each 8×8 CU, the normalized number of bits, and the maximum and minimum number of bits of the 8×8 CU from each frame in the source stream; the total number of bits used by the frame;

2) Obtain the normalized bits of each 8×8 CU, the maximum and minimum values of the bits of the 8×8 CU, and the normalization of the 8×8 CU from each frame in the target code stream The minimum number of bits; the total number of bits used in each frame is obtained from the target stream;

Find the scale factor k:

in, represents the total number of bits of the t-th frame of the target code stream, Represents the total number of bits in the t-th frame of the source stream, Indicates the maximum number of bits of the 8×8 CU in the t-th frame of the target code stream, Indicates the minimum number of bits of the 8×8 CU in the t-th frame of the target code stream, Indicates the maximum number of bits of the 8×8 CU in the t-th frame of the source stream, Indicates the minimum number of bits of the 8×8 CU of the t-th frame of the source stream;

3) Find the distribution of each 8×8 CU. The 8×8 CU that meets the following conditions must appear in the target code stream:

in Indicates the minimum number of normalized bits of the 8×8 CU in the pth CTU of the tth frame in the target code stream; Indicates the normalized number of bits of the 8×8 CU in the pth CTU of the tth frame in the source stream;

4) According to the distribution of each 8×8 CU, predict the 8×8 CUs that must appear in the target code stream except the 0th frame and the 1st frame;

5) The target code stream CTU block prediction. According to the 8×8 CUs that must appear in the target code stream, all 8×8 CUs, 16×16 CUs, 32×32 CUs, and 64×64 CUs are sequentially derived to construct a Complete CTU;

6) H.265 encoding is performed on the current frame according to the known CTU block situation.

The beneficial effect of the invention is that the information in the input code stream is fully utilized to guide H.265 encoding, the computation amount of block division and mode selection is reduced, the transcoding time is greatly shortened, and the subjectivity and objectivity of the video after transcoding are also guaranteed. visual quality. The code stream format of the present invention is not limited to H.265 isomorphic transcoding, and the technology will support H.264/AVC or H.263++, or even the heterogeneous transcoding of earlier MPEG standards.

Description of drawings

Fig. 1 is the H.265 video transcoding architecture of the present invention;

Fig. 2 is the CTU block strategy of the present invention;

Fig. 3 is the CTU construction process of the present invention;

FIG. 4 shows the distribution rules of coding modes of each frame in the embodiment period.

Detailed ways:

The video transcoding described in the present invention is mainly divided into five main steps:

The first step is the pre-processing process of decoding and transcoding. Decode the input stream to generate a YUV sequence and count the number of bits used by each CU in each frame. Analyze the header information of the input source code stream to determine the encoding standard adopted by the code stream. If it is the H.265 standard, go directly to the second step. If it is other standards, such as H.264/AVC, H.263++, MPEG-x, etc., or even a future coding standard based on block management, the bit number mapping from the standard to H.265 needs to be completed. Equation (1) is used to describe the bit number mapping relationship of the CU.

R ₂₆₅ | _QP = a ₁ · R _other ² | _QP + a ₂ · R _other | _QP + a ₃ (1)

Among them, R ₂₆₅ | _QP represents the number of bits of a CU in the H.265 code stream under the current QP, and R _other | _QP represents the bits of the corresponding CU in the encoded code stream using a coding standard other than H.265 under the specified QP The numbers, a ₁ , a ₂ , and a ₃ are three empirical parameters. In a word, the bit number mapping in heterogeneous case can be completed by formula (1). That is, the conversion of bits from other coding standards to the H.265 standard CU is realized.

The second step is the collection and processing of bit number information. Normalization of the number of bits of the target stream and statistics of the total number of bits in the frame. The first two frames of the YUV sequence are encoded with the H.265 encoder. Frame 0 is encoded in I mode, and frame 1 is encoded in P mode. The t-th frame and the number of bits used for coding of each CU in the frame are counted, and then the 8×8 CUs in the t-th frame are normalized. Count the total number of bits used by the frame. The normalization of the bit number of the source stream and the statistics of the total number of bits in the frame. That is, the 8×8 CUs in each frame in the first step are normalized using the number of bits. The normalization process is shown in formula (2).

The total number of bits used by each frame in the code stream is counted, as shown in equation (3).

w represents the number of N×N CUs in each row W represents the width of the video, H represents the height of the video, N represents the size of each LCU; h represents the number of N×N CUs in each column is the round-up operation; in formula (2), p represents the number of the CTU in the video frame, 0≤p≤w×h-1; 4·i represents the number in the CTU, i=0,1,2,… ,(N/8) ² -1; B(t,p,4·i) represents the number of bits of the 8×8 CU block of No. 4·i in the p-th block of the t-th frame; B _max (t) represents t this The maximum number of bits of a 8×8 CU block in one frame, B _min (t) represents the minimum number of bits of a CU (8×8) block in the t-th frame; in equation (3), B*(t, p,4·i) represents the number of bits of CUs of all sizes in the p-th block of the t-th frame No. 4·i, all CU blocks contain 16x16, 32x32, _64x64 , BF (t) represents the total number of bits of the t-th frame .

Step 3: 8×8 CU prediction. The presence or absence of the CU (8×8) block in the target code stream is predicted according to the normalized number of bits of the CU (8×8) block obtained by the second step. In the case of homogeneous transcoding, there is a linear relationship between the bit streams under different QPs. That is, there is a proportional relationship between the 8×8 CU normalized bits in the input code stream and the 8×8 CU normalized bits in the target code stream for the block, as shown in equation (4).

Equation (4) is transformed into Equation (5).

in Indicates the normalized number of bits of the 8×8 CU numbered 4·i in the pth CTU in the tth frame in the source stream, Indicates the normalized number of bits of the 8×8 CU numbered 4·i in the pth CTU in the tth frame in the target code stream, and k represents and ratio between. Therefore there is

in Indicates the minimum non-zero value after normalization of the number of bits of the CU block in the t-th frame in the target code stream. Therefore, when the number of normalized bits of each 8×8 CU in the source stream is greater than or equal to 1/k times the number of normalized bits of the smallest non-zero 8×8 CU in the target stream, the 8×8 The CU must appear in the target stream. According to formula (2) and formula (4), it can be deduced that

The proportional coefficient k is obtained by formula (8).

in, Indicates the number of CU bits of 8×8 number 4·i in the t-th frame p block of the source stream, Indicates the number of bits of the 8×8 CU of the t-th frame p block 4·i of the target code stream, represents the total number of bits of the t-th frame of the target code stream, Represents the total number of bits in the t-th frame of the source stream, Indicates the maximum number of bits of the 8×8 CU in the t-th frame of the target code stream, Indicates the minimum number of bits of the 8×8 CU in the t-th frame of the target code stream, Indicates the maximum number of bits of the 8×8 CU in the t-th frame of the source stream, Indicates the minimum number of bits of the 8×8 CU of the t-th frame of the source stream, is the mean of the normalized bit ratios of each block in the t-th frame of the source code stream and the target code stream.

According to Equation (2), Equation (6) and Equation (8) can predict the 8×8 CUs that must appear in the target code stream except the 0th frame and the 1st frame. The bit information of the target stream in the third step is the encoded bit information of the first frame of the YUV sequence in the second step. That is, k is obtained from the code stream of the first frame of the YUV sequence, and then used for the 8×8 CU prediction of all subsequent frames.

The fourth step, the target code stream CTU block prediction. In the third step, some 8×8 CUs are determined, and all 8×8 CU blocks are deduced according to the properties of the CTU (3); then some 16×16 CUs are deduced according to the properties of the CTU (2); The properties (3) and (2) of the CTU deduce a 32×32 CU; finally, a 64×64 CU is deduced according to the properties (3) and (2) of the CTU. Therefore, a complete CTU can be constructed from the fragment information of the 8×8 CU. If there is no 8x8 CU, the CTU only contains a large CU of 64x64. Therefore, a complete CTU can be predicted, thereby realizing the prediction of the block situation of the whole frame. Finally, the prediction of the block situation of all subsequent frames is completed.

Further, in the H.265 encoder, coding is performed according to the predicted block situation, thereby reducing the encoder's exploration of the block situation and saving encoding time. However, with the growth of the time domain, the correlation of YUV sequences in the time domain gradually weakens. In order to obtain good transcoding quality, the present invention provides a periodic transcoding scheme:

The decoded sequence can be directly coded with periodic insertion into P-frames in order to update the scale factor k. Let the coding period be ω, that is, there are ω frames in each period. The 0th frame in the cycle is coded using the I mode, the first frame is coded using the P mode, and the 2nd to ω-1th frames are first predicted and divided into blocks and then coded.

Of course, in order to further obtain good transcoding quality, some frames can be selected from the second frame to the ω-1th frame in the period and directly encoded by the H.265 encoder to update the ratio coefficient k and the minimum normalization bit. The strategy adopted: when the frame number tmodω=4 ⁿ , the H.265 encoder is directly used for encoding, and other frames are first predicted and divided into blocks and then encoded, where mod means remainder, is a round-down operation.

Example

As shown in Figure 1, fast transcoding includes:

101: Input information code stream.

102: Select an MPEG-x series decoder to decode the input code stream according to the information of the input code stream to generate a corresponding YUV file and input the YUV buffer. At the same time, count the Bit, QP, CTU, EncodeType and other information of each coding unit;

103: Select an H.264 decoder to decode the input code stream according to the information of the input code stream, and generate a corresponding YUV file to input the YUV buffer. At the same time, count the Bit, QP, CTU, encoding type EncodeType, etc. of each coding unit;

104: Select an H.265 decoder to decode the input code stream according to the information of the input code stream, and generate a corresponding YUV file to input the YUV buffer. At the same time, count the Bit, QP, CTU, encoding type EncodeType, etc. of each coding unit;

105: Receive and record the Bit, QP, CTU, EncodeType, etc. of each coding unit output by the decoder;

106: Receive and save the YUV sequence generated by the decoder;

107: judge according to the EncodeType in step 105, if EncodeType is H.265 then jump directly to step 108, if not then need to bring Bit, QP information into formula (1) to carry out the preprocessing of data to realize Bit mapping;

108: According to the result of the data preprocessing in step 107, use the three properties of the CTU to realize CTU segmentation;

109: Instruct the encoder to quickly encode the YUV through the CTU block information to obtain the final target code stream.

The flowchart of step 108 predicting the CTU block information of H.265 according to the statistical information of Bit, QP and CTU is shown in Figure 2, and the specific steps are as follows:

201: Input source code stream.

202: Obtain the number of bits of each 8×8 CU from the source code stream, and calculate the number of normalized bits and the maximum value of the normalized bits in each frame;

203: Obtain the total number of bits used in each frame from the source code stream;

204: Input the target code stream.

205: Obtain the number of bits of each 8×8 CU from the target code stream and calculate the number of normalized bits and the maximum value of the normalized bits in each frame;

206: Obtain the total number of bits used in each frame from the target code stream;

207: Bring the data of steps 203, 206, 202 and 205 into formula (8) to obtain the correlation scale coefficient k;

208: Bring the normalized bit value and the maximum value of steps 202 and 205 and the coefficient k of step 207 into formula (6) to obtain the distribution of each 8×8 CU;

209: Construct a complete CTU block tree according to the distribution of 8×8 CUs.

Step 209 in FIG. 2 describes building a CTU tree according to 8×8 CUs, and FIG. 3 shows its detailed steps.

Figure 3 depicts the construction of a CTU. Specific steps are as follows:

301: If there is no 8×8 CU, go to step 304 directly, otherwise check whether all other sibling nodes of the CU in the quadtree exist. If there is a non-existing sibling node, add it, skip to step 302;

302: Create a 16×16 CU node, check and complete its sibling nodes, and skip to step 303;

303: Create a 32×32 CU node, check and complete its sibling nodes, and skip to step 304;

304: Create a 64×64 CU node;

In the basic transcoding process shown in FIG. 1 , step 104 is to encode the decoded sequence using the H.265 encoder. In order to ensure the quality of transcoding, it needs to be encoded in cycles and the 0th frame in each cycle must be directly encoded by the H.265 encoder. In order to obtain the correlation coefficient k and the minimum number of normalized bits used in the fast transcoding process, the encoder needs to directly encode the first frame. The video's relevance decays over time. A new frame needs to be inserted at the appropriate location for the H.265 encoder to encode, to update the correlation coefficient k and the minimum number of normalization bits.

Figure 4 describes the strategy given in this embodiment. The shaded part is the frame number of the H.265 direct encoding, and the unshaded frame is the fast encoding part. The given encoding period is T=32, and the video is played at a frame rate of 30fps, that is, an update period of about 1 second.

Table 1 and Table 2 show the transcoding test performance of the method of the present invention under two conditions of homogeneous and heterogeneous. The test uses x265v1.4 as the software platform, the QP value is locked to 22, and the IPP..PP structure is encoded. The test sequence uses 9 sequences with different resolutions defined by the HEVC standard group.

The following table shows the test results of heterogeneous transcoding (H.264 stream to H.265 stream).

First encode the original YUV sequence with x264 to generate the H.264 code stream, and then use the H.264 standard decoder to reconstruct the decoded YUV sequence. The direct coding test is to directly transcode the reconstructed YUV sequence with x265v1.4, and the technical transcoding of the present invention is to quickly transcode the reconstructed YUV sequence in combination with its decoding information. From the test results, compared with the direct transcoding technology, the fast transcoding technology of the present invention achieves an average performance of 42.31% in terms of time saving, and even as high as 51.95% in the best case. At the same time, in terms of encoding quality, the fast transcoding technology is basically the same as the direct transcoding technology, with only a 0.15% loss, which can be ignored.

The following table shows the test results of isomorphic transcoding (H.265 stream to H.265 stream).

First encode the original YUV sequence with x265 to generate the H.265 code stream, and then reconstruct the decoded YUV sequence with the H.265 standard decoder. The direct coding test is to directly transcode the reconstructed YUV sequence with x265v1.4, and the technical transcoding of the present invention is to quickly transcode the reconstructed YUV sequence in combination with its decoding information. Similar to the heterogeneous transcoding test results shown in Table 1, the performance of the isomorphic transcoding test results given in Table 2 is maintained. The average performance in terms of encoding time saving reaches 44.75%, and the loss in encoding quality is only 0.89%, which is strictly controlled within 1%.

To sum up, the transcoding of the technology of the present invention can improve the coding efficiency by more than 40% while maintaining the visual quality.

Claims (3)

1. An H.265 quick transcoding method based on code stream bit number and block division mapping is characterized by comprising the following steps:

the method comprises the steps that firstly, a decoder is selected according to input source code stream information to decode a source code stream to obtain a decoding sequence, and the number of bits, quantization parameters QP and coding types of CU (coding Unit) in the source code stream are counted;

secondly, judging whether the coding type of the source code stream is H.265, if so, directly entering a third step, and if not, mapping the Bit number of the source code stream to the H.265 according to the Bit number Bit and QP and the following formula, and then entering a third step;

R₂₆₅|_QP＝a₁·R_other ²|_QP+a₂·R_other|_QP+a₃

wherein R is₂₆₅|_QPRepresents the bit number R of one CU in the H.265 code stream under the current QP_other|_QPDenotes the number of bits corresponding to CU in a code stream coded using a coding standard other than H.265 at a given QP, a₁,a₂,a₃Are three empirical parameters;

thirdly, encoding the 0 th frame of the decoding sequence by using an H.265 encoder I mode, encoding the 1 st frame by using an H.265 encoder P mode, predicting, blocking and then carrying out H.265 encoding on the rest frames;

the method for predicting the block comprises the following steps:

1) acquiring the bit number and the normalized bit number of each 8 × 8CU and the maximum value and the minimum value of the bit numbers of the 8 × 8 CUs from each frame in a source code stream; acquiring the total bit number used by each frame from the source code stream;

2) acquiring the normalized bit number of each 8 × 8CU, the maximum value and the minimum value of the bit numbers of the 8 × 8 CUs and the minimum value of the normalized bit numbers of the 8 × 8 CUs from each frame in the target code stream; acquiring the total bit number used by each frame from a target code stream;

solving a proportionality coefficient k:

wherein,representing the total bit number of the t frame of the target code stream,representing the total bit number of the tth frame of the source code stream,represents the 8 x 8CU in the t frame of the target code streamThe number of the large bits is large,represents the minimum bit number of 8 × 8CU in the t frame of the target code stream,represents the maximum bit number of 8 × 8 CUs in the tth frame of the source code stream,representing the minimum bit number of 8 multiplied by 8CU in the tth frame of the source code stream;

3) and calculating the distribution condition of each 8 × 8CU, wherein the 8 × 8CU meeting the following conditions must appear in the target code stream:

whereinExpressing the minimum value of the normalized bit number of No. 4 & i 8 multiplied by 8CU in the p-th coding tree unit CTU of the t-th frame in the target code stream;indicates the normalized bit number of the 4 & i 8 × 8CU in the p-th CTU of the t-th frame in the source code stream, 4 & i indicates the number in the CTU, i is 0,1,2,., (N/8)²-1, N denotes the size of each largest coding unit LCU;

4) predicting 8 × 8 CUs which must appear in target code streams except the 0 th frame and the 1 st frame according to the distribution condition of each 8 × 8 CU;

5) predicting target code stream CTU blocks: sequentially deducing all 8 × 8 CUs, 16 × 16 CUs, 32 × 32 CUs and 64 × 64 CUs according to the 8 × 8 CUs determined to appear in the target code stream, thereby constructing a complete CTU;

6) the current frame is h.265 encoded according to the known CTU blocking case.

2. The h.265 fast transcoding method based on bit number of code stream and block partition mapping as claimed in claim 1, wherein in the third step, the periodic transcoding is performed, in one coding period, the 0 th frame in the coding period in the decoded sequence is coded by h.265 coder I mode, the 1 st frame is coded by h.265 coder P mode, the h.265 coding is performed after the 2 nd frame to the ω -1 th frame are predicted and blocked, and ω is the total number of frames in the coding period.

3. The H.265 fast transcoding method based on bit number of code stream and block division mapping as claimed in claim 1, wherein in a coding period, a part of frames selected from the 2 nd frame to the ω -1 th frame is directly coded by using an H.265 coder, and the specific method is that when the frame number of the t-th frame satisfies tmod ω 4ⁿIn this case, the coding is performed directly using an H.265 coder, and other frames are coded after prediction blocking, where mod represents the remainder, for the rounding-down operation, ω is the total number of frames in the coding period.