CN104853191B - A Fast Encoding Method for HEVC - Google Patents

Abstract

The invention discloses a HEVC fast encoding method, which makes full use of the time-domain similarity of the largest coding unit in the encoding process and the largest coding unit with the same coordinate position in the forward reference frame of the frame where it is located, and the adjacent left side of the largest coding unit The spatial similarity of the largest coding unit and the spatial similarity of the adjacent largest coding unit above it are used to calculate the depth prediction value of the largest coding unit in the P frame or B frame, and then use the depth prediction value and depth of the largest coding unit to traverse the interval The one-to-one correspondence relationship determines the depth traversal interval of the largest coding unit, encodes the largest coding unit and each coding unit in the depth traversal interval, and encodes the largest coding unit that meets the conditions in the P frame or B frame during the encoding process And each coding unit that meets the conditions performs fast prediction mode selection, which reduces unnecessary depth and prediction mode traversal, thereby reducing the computational complexity of video coding on the premise of ensuring video quality.

Description

A Fast Encoding Method for HEVC

technical field

The present invention relates to a video coding technology, in particular to a HEVC fast coding method.

Background technique

With the rapid development of multimedia and network technology, video coding standards such as MPEG-2, MPEG-4, and H.264/AVC have been difficult to meet users' requirements for efficient compression and transmission of ultra-high-definition video. VCEG of the International Standard Organization ITU-T and MPEG of ISO/IEC jointly established JCT-VC (Joint Collaborative Team on Video Coding) to research and formulate the High Efficiency Video Coding (HEVC) standard. Compared with H.264/AVC, HEVC has basically achieved the goal of doubling the coding efficiency; however, because HEVC uses technologies such as larger coding blocks and quadtree coding structures in the coding structure, it leads to complex coding. significantly increased.

The quadtree structure division of a Largest Coding Unit (LCU) in the HEVC Test Model (HEVC Test Model, HM) needs to use a recursive method to perform a full traversal of the depth value from 0 to 3. Figure 1 shows the process of determining the final partition form of an LCU, which needs to calculate 1+4+4×4+4×4×4=85 Rate-distortion Optimization (RDO) costs, and each coding unit (Coding Unit, CU) also performs intra-frame and inter-frame traversal of various prediction unit (Prediction Unit, PU) prediction modes. Obviously, the clipping process of the entire LCU makes the computational complexity of the encoding end very large. Hou et al. used the threshold determined by the rate-distortion cost to terminate the partitioning of the CU early. This method failed to terminate the selection process of the prediction mode of the large-sized CU, so its ability to reduce complexity was very limited. Shen et al. predict the depth range (DepthRange, DR) of the current LCU by weighting the depth values of adjacent LCUs in the space-time domain, which can reduce the number of traversal depths of the LCU, but it does not consider the difference between video sequences. It uses a fixed Weights are not applicable to all video sequences, and the depth range of its prediction still needs further improvement. Xiong et al. used the optical flow method to calculate the eigenvalues of Pyramid Motion Divergence (PMD) to determine the segmentation of the CU, which reduced the complexity to a certain extent, but did not carefully consider the correlation between motion vectors, resulting in its The rate-distortion performance is not good.

Contents of the invention

The technical problem to be solved by the present invention is to provide a fast encoding method of HEVC, which can effectively reduce encoding complexity under the premise of ensuring video quality.

The technical solution adopted by the present invention to solve the above-mentioned technical problems is: a fast encoding method of HEVC, which is characterized in that it comprises the following steps:

① Define the current frame to be processed in the high-definition video as the current frame;

②Define the largest coding unit currently to be encoded in the current frame as the current largest coding unit;

③ According to the frame type of the current frame and the position of the current largest coding unit in the current frame, determine the largest coding unit for all predictions of the current largest coding unit; Defined as the prediction set of the current largest coding unit, and recorded as Ω, where Ω is an empty set or the largest coding unit contained in Ω is at least one of L, T, and COL, and L represents the left phase of the current largest coding unit Adjacent to the largest coding unit, T indicates the upper adjacent largest coding unit of the current largest coding unit, and COL indicates the largest coding unit in the forward reference frame of the current frame with the same coordinate position as the current largest coding unit;

④ Let D _pred represent the depth prediction value of the current largest coding unit, and then establish a one-to-one correspondence between the value of D _pred and the depth traversal interval of the current largest coding unit, wherein the value of D _pred does not exist or the value of D _pred is a real number , and 0≤D _pred ≤3;

⑤ If Ω is an empty set or the largest coding unit contained in Ω is one of L, T and COL or the largest coding unit contained in Ω is two of L, T and COL, then determine the value of D _pred Does not exist, then perform step ⑦; if the largest coding unit contained in Ω is L, T and COL, then determine that the value of D _pred is a real number, and 0≤D _pred ≤3, then perform step ⑥;

⑥ Calculate the value of D _pred by obtaining the spatial similarity TAS between the current largest coding unit and T, the spatial similarity LAS between the current largest coding unit and L, and the temporal similarity TS between the current largest coding unit and COL, Among them, 1≤m≤3, 1≤i≤256, and both m and i are integers, ω _m represents the weight occupied by the mth largest coding unit in Ω, when m=1, ω _m =LAS, when ω _m =TAS when m=2, ω _m =TS when m=3, Indicates the depth value of the i-th basic storage unit with a size of 4×4 in the m-th largest coding unit in Ω;

⑦ Determine the depth traversal interval of the current largest coding unit according to the one-to-one correspondence between the value of D _pred and the depth traversal interval of the current largest coding unit; Each coding unit in the unit is encoded. During the encoding process, if the frame where the current largest coding unit is located is a P frame or a B frame in high-definition video, then when the current largest coding unit or each coding unit in the current largest coding unit If the average spatio-temporal distortion value detectable by the human eye is less than the set low threshold T ₁ or greater than the set high threshold T ₂ , the fast prediction mode is performed on the current largest coding unit or each coding unit within the current largest coding unit The selection of the current largest coding unit or each coding unit in the current largest coding unit is traversed by skip, merge, inter2N×2N and intra2N×2N prediction modes, and the prediction mode with the smallest rate-distortion cost is selected as the optimal predictive model;

⑧Using the next largest coding unit to be encoded in the current frame as the current largest coding unit, and then returning to step ③ to continue until all the largest coding units in the current frame are encoded;

⑨Take the next frame to be processed in the high-definition video as the current frame, and then return to step ② to continue until all the frames in the high-definition video are processed, and the fast encoding of high-definition video is completed.

The process of determining the LCU for all predictions of the current LCU in step ③ is as follows:

If the current frame is an I frame in a high-definition video, when the current largest coding unit is the first largest coding unit in the current frame, it is determined that the current largest coding unit does not have a prediction largest coding unit; when the current largest coding unit is the current largest coding unit When there are other largest coding units in the largest coding unit in the first column of the frame except the first largest coding unit, determine the upper adjacent largest coding unit T of the current largest coding unit as the largest coding unit for prediction of the current largest coding unit; When the current largest coding unit is the remaining largest coding unit except the first largest coding unit in the first row of the largest coding unit of the current frame, determine that the left adjacent largest coding unit L of the current largest coding unit is the current largest coding unit The largest coding unit for prediction; when the current largest coding unit is the remaining largest coding unit in the current frame except the largest coding unit in the first column and the largest coding unit in the first row, determine the largest coding unit adjacent to the left of the current largest coding unit L and the upper adjacent largest coding unit T are the largest coding units used for prediction of the current largest coding unit;

If the current frame is a P frame or B frame in high-definition video, when the current maximum coding unit is the first maximum coding unit in the current frame, determine the coordinate position of the current frame's forward reference frame and the current maximum coding unit The same largest coding unit COL is the largest coding unit used for prediction of the current largest coding unit; when the current largest coding unit is the remaining largest coding unit except the first largest coding unit in the first column of the largest coding unit of the current frame, Determine that the upper adjacent largest coding unit T of the current largest coding unit and the largest coding unit COL at the same coordinate position as the current largest coding unit in the forward reference frame of the current frame are the largest coding units for prediction of the current largest coding unit; When the current largest coding unit is the remaining largest coding unit except the first largest coding unit in the first row of the largest coding unit of the current frame, determine the left adjacent largest coding unit L of the current largest coding unit and the forward direction of the current frame The largest coding unit COL at the same coordinate position as the current largest coding unit in the reference frame is the largest coding unit used for prediction of the current largest coding unit; when the current largest coding unit is the largest coding unit and the first When determining the remaining LCUs other than the LCU, determine the coordinates of the LCU L adjacent to the current LCU, the LCU T adjacent to the upper side, and the current LCU in the forward reference frame of the current frame. The same largest coding unit COL is the largest coding unit used for prediction of the current largest coding unit.

The one-to-one correspondence between the value of D _pred in step ④ and the depth traversal interval of the current largest coding unit is: when D _pred = 0, the depth traversal interval of the current largest coding unit is [0,0]; when 0 <D _pred ≤0.5, the depth traversal interval of the current largest coding unit is [0,1]; when 0.5<D _pred ≤1.5, the depth traversal interval of the current largest coding unit is [0,2]; when 1.5<D When _pred ≤ 2.5, the depth traversal interval of the current largest coding unit is [1,3]; when 2.5<D _pred ≤3, the depth traversal interval of the current largest coding unit is [2,3]; when the value of D _pred is not When it exists, the depth traversal interval of the current LCU is [0,3].

The concrete process of described step 6. is:

⑥-1. Calculate the difference between the left adjacent largest coding unit L of the current largest coding unit and the left adjacent largest coding unit L-COL of the largest coding unit at the same coordinate position as the current largest coding unit in the forward reference frame of the current frame The average depth difference between , denoted as ADD1, And calculate the average between the upper adjacent largest coding unit T of the current largest coding unit and the upper adjacent largest coding unit T-COL of the largest coding unit at the same coordinate position as the current largest coding unit in the forward reference frame of the current frame Depth difference, denoted as ADD2, Then calculate the average value of ADD1 and ADD2, denoted as ADD, ADD=(ADD1+ADD2)/2; wherein, 1≤i≤256, and is an integer, Indicates the depth value of the i-th basic storage unit whose size is 4×4 in L, Indicates the depth value of the i-th basic storage unit whose size is 4×4 in L-COL, Indicates the depth value of the i-th basic storage unit whose size is 4×4 in T, Indicates the depth value of the i-th basic storage unit whose size is 4×4 in T-COL, The value range of is [0,3], and it is an integer, and the symbol "||" is the absolute value symbol;

⑥-2. Calculate the spatial similarity between the current largest coding unit and the upper adjacent largest coding unit T of the current largest coding unit, denoted as TAS, TAS=0.05×ADD+0.25; and calculate the current largest coding unit and the current largest coding unit The spatial similarity of the adjacent largest coding unit L on the left side of , denoted as LAS, LAS=0.05×ADD+0.25; at the same time, calculate the coordinate position of the current largest coding unit and the forward reference frame of the current frame to be the same as the current largest coding unit The time-domain similarity of the largest coding unit COL of , denoted as TS, TS=-0.1×ADD+0.5;

⑥-3. Calculate the value of D _pred , Wherein, 1≤m≤3, and is an integer, ω _m represents the weight occupied by the mth largest coding unit in Ω, when m=1, ω _m =LAS, when m=2, ω _m =TAS, when When m=3, ω _m =TS, Indicates the depth value of the i-th basic storage unit with a size of 4×4 in the m-th LCU in Ω.

The specific process of encoding the current largest coding unit and each coding unit in the current largest coding unit according to the depth traversal interval of the current largest coding unit in step ⑦ is as follows:

⑦-1. Define the current coding unit to be processed in the depth traversal interval of the current largest coding unit as the current coding unit, and define the layer where the current coding unit is located as the current layer;

⑦-2. Use JND _ts to represent the average spatiotemporal distortion value of the current coding unit that can be detected by human eyes. If the current frame is an I frame in a high-definition video, determine that the value of JND _ts does not exist, and then perform step ⑦-3; If the current frame is a P frame or B frame in high-definition video, calculate the value of JND _ts , Then execute step ⑦-3, where (x, y) represents the coordinate position of the pixel in the current coding unit, 0≤x≤K-1, 0≤y≤G-1, and they are all integers, and K represents the current The total number of pixels in a row in the coding unit, G represents the total number of pixels in a column in the current coding unit, JND(x, y) represents the space-time domain of the pixel at the coordinate position (x, y) in the current coding unit Distortion values are just perceivable by the human eye;

⑦-3. When the value of JND _ts does not exist, encode the current coding unit with intra2N×2N and intraN×N prediction modes respectively, and select the prediction mode with the smallest rate-distortion cost as the optimal prediction mode of the current coding unit, Then execute step ⑦-4; when the value of JND _ts exists, judge whether the value of JND _ts is less than the set low threshold T ₁ or greater than the set high threshold T ₂ , if so, determine whether the current coding unit performs fast prediction Mode selection: use skip, merge, inter2N×2N and intra2N×2N prediction modes for the current coding unit to perform traversal coding, select the prediction mode with the smallest rate-distortion cost as the optimal prediction mode for the current coding unit, and then perform step ⑦ -4; otherwise, it is determined that the current coding unit does not select the fast prediction mode, and the current coding unit uses skip, merge, inter2N×2N, inter2N×N, interN×2N, AMP, intra2N×2N and intraN×N prediction modes respectively Perform full traversal coding, select the prediction mode with the smallest rate-distortion cost as the optimal prediction mode of the current coding unit, and then perform step ⑦-4; among them, inter2N×2N, inter2N×N, interN×2N, intra2N×2N and intraN The value of N in ×N is half of the total number of pixels in a row or a column of pixels in the current coding unit;

⑦-4. Determine whether the depth value of the current coding unit is less than the maximum value of the depth traversal interval of the current largest coding unit, and if so, further divide the current coding unit into four next-layer coding units of the same size, and then divide the The coding unit currently to be processed in the coding unit of the next layer is taken as the current coding unit, and the layer where the current coding unit is located is taken as the current layer, and then return to step ⑦-2 to continue execution; otherwise, determine that the coding process of the current coding unit is over, and then execute Step ⑦-5;

⑦-5. Determine whether all the coding units in the depth traversal interval of the current largest coding unit have been processed. If so, determine that the encoding process of the current largest coding unit is completed, and then perform step ⑧; otherwise, determine whether the four coding units in the current layer Whether all the coding units have been processed. If all the four coding units in the current layer have been processed, the next coding unit to be processed in the coding units of the upper layer of the current layer is used as the current coding unit, and the current coding unit is located Layer as the current layer, and then return to step ⑦-2 to continue execution. If the four coding units in the current layer have not been processed, take the next coding unit to be processed in the current layer as the current coding unit, and set the current coding unit where layer as the current layer, and then return to step ⑦-2 to continue.

In the step ⑦-3, T ₁ =3.5 and T ₂ =10.

Compared with the prior art, the present invention has the advantages of:

1) The method of the present invention makes full use of the time-domain similarity between the LCU and the LCU with the same coordinate position in the forward reference frame of the frame where the LCU is located in the high-definition video coding process and the LCU and the LCU. The spatial similarity of the left adjacent largest coding unit of the coding unit, the spatial similarity of the largest coding unit and the upper adjacent largest coding unit of the largest coding unit, to calculate the largest coding unit in the P frame or B frame in the high-definition video Then use the one-to-one correspondence between the depth prediction value of the largest coding unit and the depth traversal interval to determine the depth traversal interval of the largest coding unit, and in the depth traversal interval, the largest coding unit and each coding unit in the largest coding unit Encoding, during the encoding process, fast prediction mode selection is performed on the maximum coding unit that meets the conditions in the P frame or B frame and each coding unit that meets the conditions in the maximum coding unit, effectively reducing unnecessary depth traversal during the segmentation process And unnecessary prediction mode traversal in the prediction process, thereby reducing the computational complexity of video coding under the premise of ensuring video quality.

2) In the process of traversal encoding within the depth traversal interval, the method of the present invention utilizes the average spatio-temporal domain distortion value of the encoding unit to guide the rapid selection of the prediction mode, thereby reducing unnecessary traversal of the prediction mode, thereby reducing the the computational complexity of video coding.

Description of drawings

Figure 1 is a schematic diagram of the prediction and segmentation process of a LCU in the HEVC test model HM;

Fig. 2 is the overall realization block diagram of the inventive method;

Figure 3 is a maximum coding unit C and the left adjacent maximum coding unit L of the maximum coding unit C, the upper adjacent maximum coding unit T of the maximum coding unit C, and the forward reference frame of the frame where the maximum coding unit C is located The positional relationship of the largest coding unit COL that is the same as the coordinate position of the largest coding unit C;

Fig. 4a is a schematic diagram of the statistical relationship between the average spatiotemporal distortion value of the coding unit just perceivable to the human eye and the best prediction mode of the coding unit when QP=32 and the number of coding frames is 50 for the BasketballDrive (1920×1080) sequence;

Figure 4b is a schematic diagram of the statistical relationship between the average spatio-temporal distortion value of the coding unit just perceivable by human eyes and the best prediction mode of the coding unit when the sequence of BlowingBubbles (416×240) has QP=32 and the number of coding frames is 50 frames;

Figure 4c is a schematic diagram of the statistical relationship between the average spatio-temporal distortion value of the coding unit just perceivable by human eyes and the best prediction mode of the coding unit when the Traffic (2560×1600) sequence has QP=32 and the number of coding frames is 50 frames;

Figure 4d is a schematic diagram of the statistical relationship between the average spatio-temporal distortion value of the coding unit just perceivable by human eyes and the best prediction mode of the coding unit when QP=32 and the number of coding frames is 50 for the RaceHorsesC (832×480) sequence;

Fig. 5 is compared with utilizing the HM9.0 original coding method to encode the time saving percentage of 100 frames of the test sequence listed in Table 1 under the low-delay coding structure and the random access coding structure using the method of the present invention;

Figure 6a shows a comparison of the rate-distortion performance curves of encoding 100 frames of the PeopleOnStreet test sequence (2560×1600) using the method of the present invention and the original encoding method of HM9.0 under the low-delay encoding structure;

Figure 6b shows a comparison of rate-distortion performance curves for encoding Kimono test sequences (1920×1080) for 100 frames using the method of the present invention and using the HM9.0 original encoding method under the low-delay encoding structure;

Figure 6c shows a comparison of rate-distortion performance curves for encoding 100 frames of the PartyScene test sequence (832×480) using the method of the present invention and the original encoding method of HM9.0 under the low-delay encoding structure;

Figure 6d shows a comparison of the rate-distortion performance curves of encoding 100 frames of the BlowingBubbles test sequence (416×240) using the method of the present invention and the original encoding method of HM9.0 under the low-delay encoding structure;

Figure 7a is a schematic diagram of the quadtree segmentation results after encoding the sixth frame (1280×720) in the Johnny test sequence under the low-latency encoding structure and the quantization parameter QP=32 using the HM9.0 original encoding method ;

Fig. 7b is a schematic diagram of the quadtree segmentation result after encoding the sixth frame (1280×720) in the Johnny test sequence under the condition of the quantization parameter QP=32 under the low-delay encoding structure by the method of the present invention.

detailed description

The present invention will be further described in detail below in conjunction with the accompanying drawings and embodiments.

A kind of fast encoding method of HEVC proposed by the present invention, its overall realization block diagram is as shown in Figure 2, and it comprises the following steps:

① Define the current frame to be processed in the high-definition video as the current frame.

There are I frame, P frame and B frame in high-definition video coding. Among them, I frame belongs to the key frame of video coding and belongs to the I frame image group. It does not use time domain information and adopts intra frame coding; P frame adopts the way of forward reference Intra-frame and inter-frame encoding is performed, which belongs to the non-I-frame image group; B-frame uses forward and backward bidirectional reference methods for intra-frame and inter-frame encoding, and also belongs to the non-I-frame image group.

② Define the current largest coding unit to be encoded in the current frame as the current largest coding unit.

The HEVC test model HM adopts a recursive coding method from small-size coding units to large-size coding units. The sizes of coding units are 8×8, 16×16, 32×32 and 64×64. Wherein, a 64×64 coding unit is called a maximum coding unit in an encoding process.

③ According to the frame type of the current frame and the position of the current largest coding unit in the current frame, determine the largest coding unit for all predictions of the current largest coding unit; Defined as the prediction set of the current largest coding unit, and recorded as Ω, where Ω is an empty set or the largest coding unit contained in Ω is at least one of L, T and COL, that is, Ω={T} or Ω={ L} or Ω={COL} or Ω={L,T} or Ω={T,COL} or Ω={L,COL} or Ω={L,T,COL}, L represents the current largest coding unit The largest coding unit adjacent to the left, T indicates the upper adjacent largest coding unit of the current largest coding unit, and COL indicates the largest coding unit at the same coordinate position as the current largest coding unit in the forward reference frame of the current frame. Figure 3 shows a maximum coding unit C, the left adjacent maximum coding unit L of the maximum coding unit C, the upper adjacent maximum coding unit T of the maximum coding unit C, and the forward direction of the frame where the maximum coding unit C is located. The positional relationship of the largest coding unit COL at the same coordinate position as the largest coding unit C in the reference frame.

In this specific embodiment, the determination process of all prediction LCUs of the current LCU in step ③ is as follows:

If the current frame is an I frame in a high-definition video, when the current largest coding unit is the first largest coding unit in the current frame, it is determined that the current largest coding unit has no largest coding unit for prediction, that is, Ω is an empty set; when When the current largest coding unit is the remaining largest coding unit except the first largest coding unit in the first column of the largest coding unit of the current frame, determine that the upper adjacent largest coding unit T of the current largest coding unit is the prediction of the current largest coding unit The largest coding unit used, that is, Ω={T}; when the current largest coding unit is the remaining largest coding units except the first largest coding unit in the first row of the current frame, determine the current largest coding unit The left adjacent largest coding unit L is the largest coding unit for prediction of the current largest coding unit, that is, Ω={L}; when the current largest coding unit is the largest coding unit except the first column and the first row of the current frame When there are other LCUs outside the LCU, it is determined that the LCU L adjacent to the current LCU and the LCU T adjacent to the upper side are the LCUs used for prediction of the current LCU, that is, Ω={L,T }.

If the current frame is a P frame or B frame in high-definition video, when the current maximum coding unit is the first maximum coding unit in the current frame, determine the coordinate position of the current frame's forward reference frame and the current maximum coding unit The same largest coding unit COL is the largest coding unit used for prediction of the current largest coding unit, that is, Ω={COL}; when the current largest coding unit is the largest coding unit in the first column of the current frame except the first largest coding unit When determining the remaining largest coding units of the current largest coding unit, it is determined that the upper adjacent largest coding unit T of the current largest coding unit and the largest coding unit COL at the same coordinate position as the current largest coding unit in the forward reference frame of the current frame are all of the current largest coding unit The largest coding unit for prediction, that is, Ω={T,COL}; when the current largest coding unit is the remaining largest coding units in the first row of the largest coding unit of the current frame except the first largest coding unit, determine the current largest coding unit The largest coding unit L adjacent to the left of the coding unit and the largest coding unit COL at the same coordinate position as the current largest coding unit in the forward reference frame of the current frame are both the largest coding units for prediction of the current largest coding unit, that is, Ω= {L,COL}; When the current largest coding unit is the remaining largest coding unit in the current frame except the largest coding unit in the first column and the largest coding unit in the first row, determine the left adjacent largest coding unit L of the current largest coding unit The largest coding unit COL that is at the same coordinate position as the current largest coding unit in the upper adjacent largest coding unit T and the forward reference frame of the current frame is the largest coding unit for prediction of the current largest coding unit, that is, Ω={L ,T,COL}.

④ Let D _pred represent the depth prediction value of the current largest coding unit, and then establish a one-to-one correspondence between the value of D _pred and the depth traversal interval of the current largest coding unit, wherein the value of D _pred does not exist or the value of D _pred is a real number , and 0≤D _pred ≤3.

In this specific embodiment, the one-to-one correspondence between the value of D _pred in step ④ and the depth traversal interval of the current largest coding unit is: when D _pred = 0, the depth traversal interval of the current largest coding unit is [0,0 ]; when 0<D _pred ≤0.5, the depth traversal interval of the current largest coding unit is [0,1]; when 0.5<D _pred ≤1.5, the depth traversal interval of the current largest coding unit is [0,2]; When 1.5<D _pred ≤2.5, the depth traversal interval of the current largest coding unit is [1,3]; when 2.5<D _pred ≤3, the depth traversal interval of the current largest coding unit is [2,3]; when D When the value of _pred does not exist, the depth traversal interval of the current largest coding unit is [0,3].

Here, the value of D _pred needs to be calculated by weighting the average value of the depth values of the 4×4 blocks in the three largest coding units of L, T, and COL. Therefore, if Ω does not include three largest coding units, That is, if Ω is an empty set or contains one LCU or two LCUs, then the value of D _pred is considered not to exist. Therefore, when D _pred does not exist, the depth traversal interval of the current LCU is defined as [0, 3]; and if Ω contains three LCUs, then D _pred is a real number, and 0≤D _pred ≤3.

⑤ If Ω is an empty set or the largest coding unit contained in Ω is one of L, T and COL or the largest coding unit contained in Ω is two of L, T and COL, then determine the value of D _pred does not exist, then perform step ⑦; if the largest coding unit contained in Ω is L, T and COL, then determine that the value of D _pred is a real number, and 0≤D _pred ≤3, and then perform step ⑥.

⑥ Calculate the value of D _pred by obtaining the spatial similarity TAS between the current largest coding unit and T, the spatial similarity LAS between the current largest coding unit and L, and the temporal similarity TS between the current largest coding unit and COL, Among them, 1≤m≤3, 1≤i≤256, and both m and i are integers, ω _m represents the weight occupied by the mth largest coding unit in Ω, when m=1, ω _m =LAS, when ω _m =TAS when m=2, ω _m =TS when m=3, Indicates the depth value of the i-th basic storage unit with a size of 4×4 in the m-th LCU in Ω.

In this specific embodiment, the concrete process of step 6. is:

⑥-1. Calculate the difference between the left adjacent largest coding unit L of the current largest coding unit and the left adjacent largest coding unit L-COL of the largest coding unit at the same coordinate position as the current largest coding unit in the forward reference frame of the current frame The average depth difference between , denoted as ADD1, And calculate the average between the upper adjacent largest coding unit T of the current largest coding unit and the upper adjacent largest coding unit T-COL of the largest coding unit at the same coordinate position as the current largest coding unit in the forward reference frame of the current frame Depth difference, denoted as ADD2, Then calculate the average value of ADD1 and ADD2, denoted as ADD, ADD=(ADD1+ADD2)/2; wherein, 1≤i≤256, and is an integer, Indicates the depth value of the i-th basic storage unit whose size is 4×4 in L, Indicates the depth value of the i-th basic storage unit whose size is 4×4 in L-COL, Indicates the depth value of the i-th basic storage unit whose size is 4×4 in T, Indicates the depth value of the i-th basic storage unit whose size is 4×4 in T-COL, The value range of is [0,3], and it is an integer, and the symbol "||" is the absolute value symbol.

⑥-2. Calculate the spatial similarity between the current largest coding unit and the upper adjacent largest coding unit T of the current largest coding unit, denoted as TAS, TAS=0.05×ADD+0.25; and calculate the current largest coding unit and the current largest coding unit The spatial similarity of the adjacent largest coding unit L on the left side of , denoted as LAS, LAS=0.05×ADD+0.25; at the same time, calculate the coordinate position of the current largest coding unit and the forward reference frame of the current frame to be the same as the current largest coding unit The time-domain similarity of the largest coding unit COL is denoted as TS, TS=-0.1×ADD+0.5.

⑥-3. Calculate the value of D _pred , Wherein, 1≤m≤3, and is an integer, ω _m represents the weight occupied by the mth largest coding unit in Ω, when m=1, ω _m =LAS, when m=2, ω _m =TAS, when When m=3, ω _m =TS, Indicates the depth value of the i-th basic storage unit with a size of 4×4 in the m-th LCU in Ω.

⑦ Determine the depth traversal interval of the current largest coding unit according to the one-to-one correspondence between the value of D _pred and the depth traversal interval of the current largest coding unit; Each coding unit in the unit is encoded. During the encoding process, if the frame where the current largest coding unit is located is a P frame or a B frame in high-definition video, then when the current largest coding unit or each coding unit in the current largest coding unit If the average spatio-temporal distortion value detectable by the human eye is less than the set low threshold T ₁ or greater than the set high threshold T ₂ , the fast prediction mode is performed on the current largest coding unit or each coding unit within the current largest coding unit The selection of the current largest coding unit or each coding unit in the current largest coding unit is traversed by skip, merge, inter2N×2N and intra2N×2N prediction modes, and the prediction mode with the smallest rate-distortion cost is selected as the optimal predictive mode.

In this specific embodiment, the specific process of encoding the current largest coding unit and each coding unit in the current largest coding unit according to the depth traversal interval of the current largest coding unit in step ⑦ is as follows:

⑦-1. Define the current coding unit to be processed in the depth traversal interval of the current largest coding unit as the current coding unit, and define the layer where the current coding unit is located as the current layer.

⑦-2. Use JND _ts to represent the average spatiotemporal distortion value of the current coding unit that can be detected by human eyes. If the current frame is an I frame in a high-definition video, determine that the value of JND _ts does not exist, and then perform step ⑦-3; If the current frame is a P frame or B frame in high-definition video, calculate the value of JND _ts , Then execute step ⑦-3, where (x, y) represents the coordinate position of the pixel in the current coding unit, 0≤x≤K-1, 0≤y≤G-1, and they are all integers, and K represents the current The total number of pixels in a row in the coding unit, G represents the total number of pixels in a column in the current coding unit, JND(x, y) represents the space-time domain of the pixel at the coordinate position (x, y) in the current coding unit Human eyes can just detect the distortion value, and the value of JND(x, y) is obtained by using existing technology.

⑦-3. When the value of JND _ts does not exist, encode the current coding unit with intra2N×2N and intraN×N prediction modes respectively, and select the prediction mode with the smallest rate-distortion cost as the optimal prediction mode of the current coding unit, Then execute step ⑦-4; when the value of JND _ts exists, judge whether the value of JND _ts is less than the set low threshold T ₁ or greater than the set high threshold T ₂ , if so, determine whether the current coding unit performs fast prediction Mode selection: use skip, merge, inter2N×2N and intra2N×2N prediction modes for the current coding unit to perform traversal coding, select the prediction mode with the smallest rate-distortion cost as the optimal prediction mode for the current coding unit, and then perform step ⑦ -4; otherwise, it is determined that the current coding unit does not select the fast prediction mode, and use skip, merge, inter2N×2N, inter2N×N, interN×2N, AMP (Asymmetric Motion Partitioning, asymmetric motion partition) for the current coding unit respectively , intra2N×2N and intraN×N prediction modes for full traversal coding, select the prediction mode with the smallest rate-distortion cost as the optimal prediction mode of the current coding unit, and then perform steps ⑦-4; among them, inter2N×2N, inter2N×N The value of N in , interN×2N, intra2N×2N and intraN×N is half of the total number of pixels in a row or a column in the current coding unit.

Here, take T ₁ =3.5, take T ₂ =10, the specific values of T ₁ and T ₂ are determined by establishing the average spatiotemporal distortion value of the current coding unit and the optimal value of the current coding unit. The statistical relationship between the prediction modes (that is, the prediction mode with the smallest rate-distortion cost value) is determined by using the statistical relationship.

Since the prediction mode of video coding varies with the region and the degree of difference between the current frame and the reference frame, the human-eye just perceptible distortion model indicates the maximum distortion that the human eye cannot perceive, including the spatial domain human-eye just perceptible distortion model and temporal The human eye is just perceptible to the human eye in the domain domain, and the distortion model in the spatial domain is just perceptible to the human eye, which shows that the human eye is more sensitive to relative brightness than absolute brightness. The texture mask shows that textured areas can hide more distortion than flat areas; The perceivable distortion model shows that the greater the difference between the two frames before and after, the greater the distortion that can be hidden. Therefore, the selection of the inter-frame prediction mode can be guided by the perceptible distortion model of human eyes. Figure 4a shows a schematic diagram of the statistical relationship between the average spatio-temporal distortion value of the encoding unit and the best prediction mode of the encoding unit when the quantization parameter is equal to 32 and the number of encoding frames is 50 for the BasketballDrive sequence; Figure 4b shows A schematic diagram of the statistical relationship between the average spatio-temporal distortion value of the coding unit just perceptible to the human eye and the best prediction mode of the coding unit when the quantization parameter is equal to 32 and the number of coding frames is 50 for the BlowingBubbles sequence is shown; When the quantization parameter is equal to 32 and the number of encoded frames is 50 frames, the statistical relationship between the average spatio-temporal distortion value of the encoding unit and the best prediction mode of the encoding unit is shown; Figure 4d shows the quantization of the RaceHorses sequence When the parameter is equal to 32 and the number of encoding frames is 50 frames, it is a schematic diagram of the statistical relationship between the average spatio-temporal distortion value of the coding unit just perceptible to the human eye and the best prediction mode of the coding unit. According to the statistical analysis in Fig. 4a to Fig. 4d, when the distortion value can be detected by human eyes in the time-space domain is small or large, the proportion of some prediction modes is very small, and these prediction modes can be excluded. Therefore, in the present invention, when the value of JND _ts is less than the set low threshold T ₁ or greater than the set high threshold T ₂ , the current coding unit selects the fast prediction mode, skip, merge, inter2N×2N and intra2N ×2N prediction mode is traversed and encoded, and an optimal prediction mode is selected.

⑦-4. Determine whether the depth value of the current coding unit is less than the maximum value of the depth traversal interval of the current largest coding unit, and if so, further divide the current coding unit into four next-layer coding units of the same size, and then divide the The coding unit currently to be processed in the coding unit of the next layer is taken as the current coding unit, and the layer where the current coding unit is located is taken as the current layer, and then return to step ⑦-2 to continue execution; otherwise, determine that the coding process of the current coding unit is over, and then execute Step ⑦-5.

⑦-5. Determine whether all the coding units in the depth traversal interval of the current largest coding unit have been processed. If so, determine that the encoding process of the current largest coding unit is completed, and then perform step ⑧; otherwise, determine whether the four coding units in the current layer Whether all the coding units have been processed. If all the four coding units in the current layer have been processed, the next coding unit to be processed in the coding units of the upper layer of the current layer is used as the current coding unit, and the current coding unit is located Layer as the current layer, and then return to step ⑦-2 to continue execution. If the four coding units in the current layer have not been processed, take the next coding unit to be processed in the current layer as the current coding unit, and set the current coding unit where layer as the current layer, and then return to step ⑦-2 to continue.

⑧ Use the next largest coding unit to be encoded in the current frame as the current largest coding unit, and then return to step ③ to continue until all the largest coding units in the current frame are encoded.

⑨Take the next frame to be processed in the high-definition video as the current frame, and then return to step ② to continue until all the frames in the high-definition video are processed, and the fast encoding of high-definition video is completed.

The method of the present invention is tested below to further illustrate the effectiveness and feasibility of the method of the present invention.

The inventive method selects HM9.0 of HEVC as the test model; the hardware configuration is a CPU intel core i5-2500, a PC with a main frequency of 3.30GHz and a memory of 8G, the operating system is Windows7, and the development tool is Microsoft VisualStudio 2008. Adopt low-latency and random access coding structure, select all test sequences to test under the four conditions of quantization parameters QP 22, 27, 32 and 37, the number of test frames is 100 frames, and the non-I frame image group contains 8 non-I frames frame, the I frame period is 32. Using BD-PSNR ( delta peak signal-to-noise rate), BDBR ( deltabit rate) and ΔT as the evaluation index of the method of the present invention. BD-PSNR represents the amount of change in PSNR under the same code rate conditions, BDBR represents the percentage change in code rate under the same PSNR conditions, ΔT(%) is used to represent the percentage change in encoding time, Wherein, T _p and T _HM correspond to represent the encoding time of adopting the method of the present invention and the HM9.0 algorithm.

Table 1 shows the test sequences selected by the method of the present invention. Table 2 shows the coding performance of 100 frames encoded by the method of the present invention to the test sequences listed in Table 1, that is, the values of BD-PSNR, BDBR and ΔT. As can be seen from Table 2, the method of the present invention has a low latency Under the temporal coding structure, BDBR only increased by 1.71%, BD-PSNR decreased by only 0.058dB, and the time was saved by 41.35%; under the random access coding structure, BDBR was only increased by 1.06%, BD-PSNR was only decreased by 0.037dB, and the time was saved by 41.43%; Compared with HM9.0, the method of the present invention basically keeps the video quality unchanged no matter in the low-delay coding structure or the random access coding structure, but greatly reduces the computational complexity of video coding.

Fig. 5 has provided and compared with using the HM9.0 original coding method to encode the test sequence listed in Table 1 under the low-latency coding structure and the random access coding structure by using the method of the present invention, the time saving percentage situation diagram of 100 frames is shown . It can be seen from Figure 5 that the average time saving ratios under the two encoding structures are similar, both reaching over 40%. Therefore, the method of the present invention reduces the computational complexity of encoding very obviously.

Figure 6a shows a comparison of rate-distortion performance curves for encoding 100 frames of the PeopleOnStreet test sequence (2560×1600) using the method of the present invention and the original encoding method of HM9.0 under a low-delay encoding structure; Figure 6b provides Using the method of the present invention and utilizing the HM9.0 original encoding method to encode the Kimono test sequence (1920 × 1080) under the low-delay encoding structure, the rate-distortion performance curve comparison diagram of 100 frames; Fig. Using the HM9.0 original encoding method to encode the PartyScene test sequence (832×480) under the low-delay encoding structure, the comparison chart of the rate-distortion performance curve of 100 frames; The encoding method encodes the BlowingBubbles test sequence (416×240) under the low-latency encoding structure and compares the rate-distortion performance curves of 100 frames. It can be seen from Figures 6a to 6d that the rate-distortion performance curves of the method of the present invention and the original encoding method of HM9.0 basically overlap, indicating the video quality after encoding by the method of the present invention and the original encoding method of HM9.0 remain largely unchanged.

Figure 7a shows the quadtree segmentation after encoding the sixth frame (1280×720) in the Johnny test sequence using the HM9.0 original encoding method under the low-latency encoding structure and the quantization parameter QP=32 Result schematic diagram; Fig. 7 b has provided the quadtree division after the 6th frame (1280 * 720) in the Johnny test sequence is encoded under the condition of quantization parameter QP=32 by using the method of the present invention under the low delay coding structure Schematic diagram of the results. The black boxes in FIG. 7a and FIG. 7b represent the differences between the two coding unit divisions, and the black vertical or horizontal lines represent the differences between the two prediction modes. It can be seen from Fig. 7a and Fig. 7b that compared with the original encoding method of HM9.0, the segmentation depth of the coding unit and the prediction mode of the prediction unit are basically the same in the method of the present invention. The split depth is similar to the prediction mode of the PU. Therefore, it can be fully demonstrated that the overall rate-distortion performance of encoding using the method of the present invention remains stable.

The test sequence that table 1 method of the present invention selects for use

sequence name resolution number of frames frame rate bit depth Traffic 2560×1600 150 30fps 8 PeopleOnStreet 2560×1600 150 30fps 8 Kimono 1920×1080 240 24fps 8 Cactus 1920×1080 500 50fps 8 ParkScene 1920×1080 240 24fps 8 BasketballDrive 1920×1080 500 50fps 8 BQ Terrace 1920×1080 600 60fps 8 PartyScene 832×480 500 50fps 8 Race Horses C 832×480 300 30fps 8 Basketball Drill 832×480 500 50fps 8 BQMall 832×480 600 60fps 8 Race Horses 416×240 300 30fps 8 Blowing Bubbles 416×240 500 50fps 8 Basketball Pass 416×240 500 50fps 8 Vidyo1 1280×720 600 60fps 8 Vidyo3 1280×720 600 60fps 8 Vidyo4 1280×720 600 60fps 8 Johnny 1280×720 600 60fps 8 FourPeople 1280×720 600 60fps 8

Kristen And Sara 1280×720 600 60fps 8

Table 2 adopts the inventive method to encode the coding performance situation of 100 frames to the test sequence listed in table 1

Claims (5)

1. a kind of HEVC fast encoding method, it is characterised in that comprise the following steps：

1. it is present frame by frame definition currently pending in HD video；

2. current maximum coding unit to be encoded in present frame is defined as current maximum coding unit；

3. the position according to the frame type and current maximum coding unit of present frame in the current frame, it is determined that current maximum coding is single The maximum coding unit of all predictions of member；Then will be single by the maximum coding of all predictions of current maximum coding unit The set that member is constituted is defined as the prediction sets of current maximum coding unit, and is designated as Ω, wherein, Ω in empty set or Ω by wrapping The maximum coding unit contained is at least one in L, T and COL, and L represents the adjacent maximum volume in the left side of current maximum coding unit Code unit, T represents the adjacent maximum coding unit in the top of current maximum coding unit, and COL represents the forward reference frame of present frame In coordinate position identical maximum coding unit with current maximum coding unit；

4. D is made_predThe depth prediction value of current maximum coding unit is represented, D is then established_predValue and current maximum coding it is single The interval one-to-one relationship of the extreme saturation of member, wherein, D_predValue be not present or D_predValue be real number, and 0≤D_pred≤ 3；

If 5. Ω is that the maximum coding unit included in empty set or Ω is included in one or Ω in L, T and COL Maximum coding unit is two in L, T and COL, it is determined that D_predValue be not present, then perform step 7.；If institute in Ω Comprising maximum coding unit be L, T and COL, it is determined that D_predValue be real number, and 0≤D_pred≤ 3, then perform step 6.；

6. by obtaining current maximum coding unit and T spatial domain similarity TAS, current maximum coding unit and L spatial domain phase Like degree LAS, current maximum coding unit and COL time domain similarity TS, D is calculated_predValue,Wherein, 1≤m≤3,1≤i≤256, and m and i are integer, ω_mRepresent in Ω M-th of maximum coding unit shared by weight, the ω as m=1_m=LAS, the ω as m=2_m=TAS, the ω as m=3_m= TS,Represent depth of i-th of the size in m-th of maximum coding unit in Ω for 4 × 4 basic unit of storage Angle value；

The detailed process of described step 6. is：

6. -1, calculate in the adjacent maximum coding unit L in the left side of current maximum coding unit and the forward reference frame of present frame with Between the adjacent maximum coding unit L-COL in the left side of the coordinate position identical maximum coding unit of current maximum coding unit Mean depth is poor, is designated as ADD1,And calculate the upper of current maximum coding unit In the forward reference frame of the adjacent maximum coding unit T in side and present frame with the coordinate position identical of current maximum coding unit most Mean depth between the adjacent maximum coding unit T-COL in top of big coding unit is poor, is designated as ADD2,Then ADD1 and ADD2 average value is calculated, ADD, ADD=(ADD1+ is designated as ADD2)/2；Wherein, 1≤i≤256, and be integer,Represent the basic storage that i-th of size in L is 4 × 4 The depth value of unit,Depth value of i-th of the size in L-COL for 4 × 4 basic unit of storage is represented,Depth value of i-th of the size in T for 4 × 4 basic unit of storage is represented,Represent in T-COL The depth value for the basic unit of storage that i-th of size is 4 × 4, Span be [0,3], and for integer, symbol " | | " is the symbol that takes absolute value；

6. current maximum coding unit maximum coding unit T adjacent with the top of current maximum coding unit spatial domain -2, is calculated Similarity, is designated as TAS, TAS=0.05 × ADD+0.25；And calculate current maximum coding unit and current maximum coding unit The adjacent maximum coding unit L in left side spatial domain similarity, is designated as LAS, LAS=0.05 × ADD+0.25；Meanwhile, calculate currently most Coordinate position identical maximum coding in the forward reference frame of big coding unit and present frame with current maximum coding unit is single First COL time domain similarity, is designated as TS, TS=-0.1 × ADD+0.5；

6. D -3, is calculated_predValue,Wherein, 1≤m≤3, and be integer, ω_mTable Show the weight shared by m-th of maximum coding unit in Ω, the ω as m=1_m=LAS, the ω as m=2_m=TAS, as m=3 ω_m=TS,Represent the basic storage list that i-th of size in m-th of maximum coding unit in Ω is 4 × 4 The depth value of member；

7. according to D_predValue and current maximum coding unit the interval one-to-one relationship of extreme saturation, it is determined that current maximum The extreme saturation of coding unit is interval；Then it is interval single to current maximum coding according to the extreme saturation of current maximum coding unit Each coding unit in member and current maximum coding unit is encoded, in an encoding process, if current maximum coding is single Frame where member is P frames or B frames in HD video, then when every in current maximum coding unit or current maximum coding unit The average time-space domain human eye of individual coding unit just perceives the Low threshold T that distortion value is less than setting₁Or more than the high threshold of setting T₂, then the choosing of fast prediction pattern is carried out to each coding unit in current maximum coding unit or current maximum coding unit Select, to each coding unit in current maximum coding unit or current maximum coding unit respectively with skip, merge, Inter2N × 2N and intra2N × 2N predictive modes carry out traversal coding, choose the minimum predictive mode of rate distortion costs value and make For optimal prediction modes；

8. 3. next maximum coding unit to be encoded in present frame, as current maximum coding unit, is then back to step Continue executing with, until all maximum coding units in present frame are encoded and finished；

9. the pending frame of next frame in HD video is then back to step and 2. continued executing with as present frame, until high definition All frames in video are disposed, and so far complete the fast coding of HD video.

2. a kind of HEVC according to claim 1 fast encoding method, it is characterised in that described step 3. in it is current The determination process of the maximum coding unit of all predictions of maximum coding unit is：

If present frame is the I frames in HD video, the 1st maximum coding in current maximum coding unit is present frame During unit, it is determined that the maximum coding unit that current maximum coding unit is not previously predicted；When current maximum coding unit is current In 1st row maximum coding unit of frame during remaining maximum coding unit in addition to the 1st maximum coding unit, it is determined that current maximum The adjacent maximum coding unit T in top of coding unit is the maximum coding unit of the prediction of current maximum coding unit；When work as Preceding maximum coding unit is compiled for remaining maximum in the 1st row maximum coding unit of present frame in addition to the 1st maximum coding unit During code unit, it is determined that the adjacent maximum coding unit L in the left side of current maximum coding unit is the prediction of current maximum coding unit Maximum coding unit；Except the 1st row maximum coding unit and the 1st row maximum are compiled in current maximum coding unit is present frame During remaining maximum coding unit outside code unit, it is determined that the adjacent maximum coding unit L in the left side of current maximum coding unit and upper The adjacent maximum coding unit T in side is the maximum coding unit of the prediction of current maximum coding unit；

If present frame is the P frames or B frames in HD video, the 1st in current maximum coding unit is present frame is most During big coding unit, determine that the coordinate position identical maximum in the forward reference frame of present frame with current maximum coding unit is compiled Code unit COL is the maximum coding unit of the prediction of current maximum coding unit；When current maximum coding unit is present frame The 1st row maximum coding unit in remaining maximum coding unit in addition to the 1st maximum coding unit when, maximum compiled it is determined that current Coordinate bit with current maximum coding unit in the code adjacent maximum coding unit T in top of unit and the forward reference frame of present frame Put the maximum coding unit that identical maximum coding unit COL is the prediction of current maximum coding unit；When current maximum Coding unit is remaining maximum coding unit in the 1st row maximum coding unit of present frame in addition to the 1st maximum coding unit When, it is determined that in the current adjacent maximum coding unit L in the left side of maximum coding unit and the forward reference frame of present frame with it is current most The coordinate position identical maximum coding unit COL of big coding unit is the maximum volume of the prediction of current maximum coding unit Code unit；In current maximum coding unit is present frame in addition to the 1st row maximum coding unit and the 1st row maximum coding unit During remaining maximum coding unit, it is determined that the adjacent maximum coding unit L in the left side and the adjacent maximum in top of current maximum coding unit With the coordinate position identical maximum coding unit of current maximum coding unit in the forward reference frame of coding unit T and present frame COL is the maximum coding unit of the prediction of current maximum coding unit.

3. a kind of HEVC according to claim 1 or 2 fast encoding method, it is characterised in that described step 4. in D_predValue and the interval one-to-one relationship of extreme saturation of current maximum coding unit be：Work as D_predIt is current maximum when=0 The extreme saturation interval of coding unit is [0,0]；When 0<D_predWhen≤0.5, the extreme saturation of current maximum coding unit is interval For [0,1]；When 0.5<D_predWhen≤1.5, the extreme saturation interval of current maximum coding unit is [0,2]；When 1.5<D_pred≤ When 2.5, the extreme saturation interval of current maximum coding unit is [1,3]；When 2.5<D_predWhen≤3, current maximum coding unit Extreme saturation interval be [2,3]；Work as D_predValue when being not present, the extreme saturation interval of current maximum coding unit for [0, 3]。

4. a kind of HEVC according to claim 1 fast encoding method, it is characterised in that described step 7. middle basis Interval each in current maximum coding unit and current maximum coding unit of the extreme saturation of current maximum coding unit The detailed process that coding unit is encoded is：

7. -1, by the extreme saturation of current maximum coding unit it is interval in currently pending coding unit be defined as present encoding Unit, current layer is defined as by the layer where current coded unit；

7. -2, JND is used_tsRepresent that the average time-space domain human eye of current coded unit just perceives distortion value, if present frame is height I frames in clear video, it is determined that JND_tsValue be not present, then perform step 7. -3；If present frame is the P in HD video Frame or B frames, then calculate JND_tsValue,Then step is performed 7. -3, wherein, (x, y) is represented The coordinate position of pixel in current coded unit, 0≤x≤K-1,0≤y≤G-1, and be all integer, K represents present encoding The total number of one-row pixels point in unit, G represents the total number of a row pixel in current coded unit, and JND (x, y) represents to work as Coordinate position just perceives distortion value for the time-space domain human eye of the pixel of (x, y) in preceding coding unit；

7. -3, JND is worked as_tsValue when being not present, mould is predicted with intra2N × 2N and intraN × N respectively to current coded unit Formula is encoded, and chooses the minimum predictive mode of rate distortion costs values as the optimal prediction modes of current coded unit, then Perform step 7. -4；Work as JND_tsValue in the presence of, judge JND_tsValue whether be less than setting Low threshold T₁Or more than setting High threshold T₂, if it is, judging that current coded unit carries out the selection of fast prediction pattern, current coded unit is used respectively Skip, merge, inter2N × 2N and intra2N × 2N predictive modes carry out traversal coding, choose rate distortion costs value minimum Predictive mode as the optimal prediction modes of current coded unit, then perform step 7. -4；Otherwise, it is determined that present encoding list Member uses skip, merge, inter2N × 2N, inter2N respectively without the selection of fast prediction pattern to current coded unit × N, interN × 2N, AMP, intra2N × 2N and intraN × N predictive modes carry out full traversal coding, selection rate distortion generation Minimum predictive mode is worth as the optimal prediction modes of current coded unit, step 7. -4 is then performed；Wherein, The value of N in inter2N × 2N, inter2N × N, interN × 2N, intra2N × 2N and intraN × N is present encoding list The half of the total number of one-row pixels point or a row pixel in member；

7. -4, judge whether the depth value of current coded unit is less than the interval maximum of extreme saturation of current maximum coding unit Value, if it is, current coded unit is divided further into next layer of coding unit of four size identicals, then will Currently pending coding unit is as current coded unit in the next layer of coding unit, by the layer where current coded unit As current layer, return again to step and 7. -2 continue executing with；Otherwise, it determines current coded unit cataloged procedure terminates, then perform Step 7. -5；

7. -5, judge whether all coding units in the extreme saturation interval of current maximum coding unit are disposed, such as It is really, it is determined that current maximum coding unit cataloged procedure terminates, and then performs step 8.；Otherwise, four in current layer are judged Whether individual coding unit is disposed, if four in current layer coding unit is disposed, by the upper of current layer Next pending coding unit is as current coded unit in one layer of coding unit, and by the layer where current coded unit As current layer, it is then back to step and 7. -2 continues executing with, finished if four in current layer coding unit is untreated, will Next pending coding unit is as current coded unit in current layer, and using the layer where current coded unit as working as Front layer, is then back to step and 7. -2 continues executing with.

5. a kind of HEVC according to claim 4 fast encoding method, it is characterised in that 7. described step takes T in -3₁ =3.5, take T₂=10.