Name: H.265 and H.264 Compared to Analysis: Yinsong Time: June 14, 2017
content
I. H.265 (HEVC) and H.264 Comparison II, H.265 Key Technical Analysis of H.265 and H.264 Coding Algorithm Comparison and Performance Test Analysis
** H.265 and H.264 comparison analysis **
1
ITU officially approved HEVC / H in early 2013. HigheffiCyvideo Coding, with H. 264 / AVC comparison, in the same video quality and application conditions, the Code rate of HEVC is nearly half, which means that the content provider can transmit higher quality network video, saving corresponding network bandwidth and data storage expenses. This article is H. 264 / h. The key technology of 265 is reviewed, and the research and analysis of the different characteristics are compared. I. H.265 (HEVC) Compared with the H.264 compared to the H.264, it does not change the original technology on the basis of H.264 but improves some related technologies. The application new technology has improved the code stream, coding quality, and delay algorithm complex. The relationship between the degree is to obtain the optimal setting. The main research content is: enhance compression efficiency, robustness and error recovery ability, reduce real-time delay, channel acquisition time, random access delay, complexity, and so on. The H.264 can only be transmitted to the standard-cleaned digital image at a speed of less than 1 Mbps, but H.265 can transmit 720P (resolution) general high-definition video at a speed of from 1 to 2 Mbps. From the storage cost, H.264 can save approximately 50% of storage space and significantly reduce the storage cost of the video. Compared with the maximum advantage of H.264, H.265, it is possible to reduce the data transmission bandwidth to 50% of H.264 under conditions that ensure that the picture quality is basically invariant, and it also supports the distinguishing of as high as 7680 × 4320 Rate. So even if the H.265 can be encoded using H.265 even if the H. 265 can be encoded. In terms of encoding architecture, H.265 is similar to H.264, mainly composed of frame prediction, inter prediction, conversion, quantization, demultiple block filter, entropy encoding, etc. The H.264 encoding architecture is divided into three basic units: coding unit, predictive unit, and conversion unit. In the coding unit, the size of the macco (MB) in H.264 is fixed to pixels, but H.265 can be selected from it. In addition, H.265 also uses updated motion vector tests, completing more improvements in multi-core paralling work, and under certain new technologies, compression processing can be efficiently completed. II. H.265 Key Technologies Analysis H.265 The core coding module includes: intra prediction module, inter prediction module, transformation, and quantization module, a semantic entropy encoding module, an adaptive annular filter module, and the like. Determining the image is determined by the encoder to determine the image using an intra or inter prediction method. H. The encoding frame of 265 / hevc is shown below. Inter Bloom prediction compressible video time domain redundancy, intra predictive compressible video hollow domain redundancy. The transform technique in the frame can further compress the airspace redundancy, thereby enabling the signal from a small amount of feature vector linear combination. Statistical redundancy in compressed coding parameters based on semantic entropy encoding technology. 1. The input image is divided into CTU. 2. The first frame image can only perform intra prediction as an I frame. The intra prediction is based on the reconstruction unit. Therefore, the input image will directly enter the quantization and inverse quantity, the part reconstruction pixels in the decoded image are used to perform intra prediction. 3. The intra prediction is completed. The encoder controller hits the switch to the intra predicted output port, and the output predictive image acquires the residual as the original image. The information in intra prediction will be transmitted to the CABAC. 4. The residual residual generates compression information by transform and quantify. This part of the information will also be transferred to the CABAC. At the same time, for the next frame of frame asking, it is necessary to quantify the quantized image residual reflexion to the original image residual, and then superimpose to form a complete decoded image in accordance with the intra predicted image, then perform the later filtering processing and SAO processing. . After this. The post-processing information will be transferred to the CABAC. 5. The original image of the second input will perform motion estimation with the decoded image, and the resulting motion vector is used to perform motion compensation, and the encoding controller hits the switch to the output port of the motion compensation. Then. The compensation image will reduce the residual processing with the original image. And transfer to step 4. The motion estimation and motion compensation information at this time will be transmitted to the CABAC for entropy encoding. 6. Repeat L ~ 5 steps. For intra prediction and frame prediction, the encoding controller controls the switching of the switch.
2.1 Encoder Basic Data Structure H. The coded tree unit CTU (Coding Tree Unit), the prediction unit PU (PREDICTION UNIT), and Transform Unit) are defined in the 265 / I-IEVC. The CTU consists of a plurality of Cu, and the CU can be divided into a plurality of PUs, the Cu can be divided into multiple TUs when the Cu may be divided into multiple PUs, and the Cu may be divided into multiple TUs. The corresponding pixel blocks in these coded block units are referred to as encoding trees CTB (Coding Tree Block), the Coding Block, and the PB (TransformLock). Image split into CTUs that are not overlapping each other, which is similar to macro blocks in H.264, but in H.265 / HEVC. The size of the CTU can be 8x8, 16x16, 32x32 or 64x64. These changes have enabled H.265 / HEVC to determine the size of the coding unit based on the amount of information in different regions of the image, compared to H.264 more flexible and efficient. The CTU is divided into multiple Cu by four-tree delivery method. The basic unit encoded in H.265 / HEVC is the encoding of each leaf node, which is used to indicate the prediction mode of the current block as an intra or frame. The size of the Cu is the size of the CTU, minimum is 8x8. When CU has multiple layers, you should traverse each leaf node in accordance with the principle of depth priority.
2.2 Prediction and transform unit H. The smallest unit of the prediction for intra or frames in the frame is PU 265 / I-IEVC. The PU contains intra prediction information for the direction predicted, or a frame prediction information for selecting a reference frame and motion vector. H. 8 PU divided types are defined in 265, as shown in Figure 2. The PU division type according to the intra predicted should be 2nx2n or NxN; the PU division type predicted as an inter prediction can be arbitrary. The asymmetric four modes in Figure 2 are optional. Can be used by the encoder configuration to be used. TU is a unit having the same residual factor or transform coefficient when an integer transformation or quantization is performed. TU can be divided according to the square division. When the Tu is square, its size can be from 4 x4 to 32 × 32. The Cu division in Figure 3 is represented by a solid line, and the TU divided by a broken line. When the TU is rectangular. Can be divided in the vertical and horizontal direction. But regardless of the TU to choose which division method, a quadruple manner can be used, such a unified data structure makes H. 265 / hevc encoding efficiency ratio H. 264 / AVC has greatly improved. 2.3 Loop filtering technology H.265 / HEVC loop filtering technology is to generate reconstruction reference frames in the encoder. H.265 / i-IEVC loop filtering includes deck filter, adaptive sample compensation SAO (Sample Adaptive Offset) and adaptive loop filter ALF (AdaptiveLoop Filtero's main role is to reduce PU during prediction Edge effect. Although H.264 also uses the same deck filter, the block filter is used in almost all 8 x 8 × 8 pieces of block edges in H.265 / HEVC, and 4 × 4-size The edge of the transform block does not filter. HEVC is different from the H.26 劬 WC standard, and the H.265 standard decoding increases the adaptive sample compensation filter SAO.SAO to increase the appropriate value for each sampling value on the ground filter. Offset to obtain more accurate sampling data, the operation only needs to be completed by a simple query index. The adaptive loop filter ALF in H.265 is used to remove filtering and SAO signal reconstruction. ALF is also another new technique proposed in the loop filter. The image coding coefficient is transmitted by bitstream. The receiving end can use a specific coefficient of two-dimensional widopic filter according to the omalection of the reconstructed image. The reconstructed image is as close as possible to the original image. 2.4 Inframe prediction and inter-frame prediction intra prediction technique started from H.264 / AVC. Used to remove spatial redundancy between adjacent blocks, thus To a higher compression efficiency. H.265 / hevc compares the previous generation H.264 / AVC intra predictive code, increasing the prediction direction and improves the intra prediction mode selection algorithm. There are 9 kinds of inhiba.264 / AVC The prediction mode includes one DC prediction mode and eight directional prediction modes, and 35 prediction modes are defined in H.265 / HEVC, including planar mode, mean prediction mode, and 33 directional prediction modes. Prediction direction The increase in numbers makes the prediction of more refined, the new planar prediction mode can effectively perform intra prediction of the area having a certain texture characteristic, thereby effectively reducing redundant improvement accuracy. 2.5 Entropy coding technology compared to H.264 H.265 / HEVC does not have a simple continuation of CAVLC and CABAC, but is improved separately, and propose low complexity entropy encoding LCEC (Low Complexity Entropy and High Complex Entropy Coding HCEC (High Complexity Entropy) The concept LCEC entropy coding scheme uses table structured and index, structured processing of the original variable length coding scheme. Different code tables are specified according to different probability distributions. Becec entropy coding scheme is based on grammar Parallel adaptive binary arithmetic coding scheme for elements, by reducing codewords encoded in context. Reduce context model selection dependence. To reduce the dependence, dependent, The demand of storage space. H.265 increases the throughput of entropy encoders. The parallelization of the encoder is optimized, so it is more suitable than the H.264 / AVC more suitable for real-time transmission of video at high resolution. III. H.265 and H.264 Coding Algorithm Comparison and Performance Test Analysis 3.1 H.265 The H.264 Coding Technology Compare H.265 has a larger macroblock. Maximum support for 64 × 64 pixels, while H.264's maximum support is 16 × 16; H.265's intra prediction method has 35 species, while H.264 is only 10 prediction methods; H.265 support The time domain + airspace vector prediction method, while H.264 only supports a vector prediction method of airspace; H.265 supports an 8-order interpolation filter. H.264 only supports 6-order jam filter; HEVC supports 8 × 8 / 16x 16/32 × 32 DCT transformation. The H.264 only supports 4x418x8 DCT transformation. 3.2 Test Environment 1 Encoder Selection: H.264 / MPEG-AVC Encoder: For the Test of the H.264 / MPEG-AVC standard, open source X264, using the latest version released by the official website of X264 on April 22, 2014. The x264 begins in 2003, from open source communities, it is an efficient, high-speed and reliable video encoder. Since it is very good in the trade-off efficiency and computational complexity, it has been popular in the network. Now, x264 is one of the most popular and popular encoders in the Internet. H.265 / MPEG-HEVC encoder: x265 is a free open source HEVC coding library, which is developed with the X264 encoder. The x265 allows all users to use under the GNU GPL 2 license agreement. Based on X264's excellent performance, we use X265 as a representative encoder of HEVC, and select the latest version 1.1. At the same time, in the coding speed and CPU usage, the 2.0 and 3.0 version of the commercial HEVC encoder are also used as the auxiliary test. 2 test environment configuration: CPU: Intel i5 3470, quad core four-thread, main frequency 3.2GHz memory: 4G DDR3 1600 operating system: Windows 7 64bit 3 Test standards and test material encoders encoding quality by calculating PSNR. The parameter setting is shown in Table. PSNR (peak signal to noise ratio) is an objective standard for evaluating images, in db (decibels). The usual image is typically between 30 to 40 dB, the higher the PSNR, the better the encoded image quality. The test material is selected from the 5-segment 1920 × 1080 resolution video sequence and 3 segment 720 × 576 resolution video sequence, each of which frame rates of each video sequence are 25 frames. In order to save the test time, each test sequence is only 200 frames, which is the length of time in 8 seconds. 5 paragraphs 1920 × 1080 video sequences are American songs and dances, Chinese basketball, military battles, leisure time and news propaganda films, 3 sections 720 × 576 video are zoo, news, movies. The use of the code stream comparison benchmark: experiments test the PSNR after each video sequence 1m to 6m code stream. The 1920 × 1080 test sequence is based on the encoded stream 5m; the 720 × 576 test sequence is based on the encoded stream 2m. Because the test data between the material with material will fluctuate, we choose the average of multiple test material results to draw the final conclusion. 4 Test Tools: VQMT (Video Quality Measurement Tool "is a video quality assessment tool, which is a third-party organization based on an OpenCV open source library, which can be supported.A variety of raw image sequences such as YUV420, YUV444 and YUV422 are suitable. It is a free video quality evaluation software that tests a number of major video quality indicators including PSNR, SSIM, and is easy to operate and easy to use. 3.3 Test Results 1H.265 Test Results with H.264 H.265 Comparison with H.264, we choose to compare at the default speed, because x265 and x264 are homologous encoders, so it is default The parameter is also closest; the encoding quality of X264 is far behind the code quality at its default speed, and cannot reflect its true coding quality. Based on the above two reasons, we chose the default speed of the two encoders to encode testing, and the test results of multiple sequences are shown in Figure 1. As shown in Figure 2, the test results of the six video sequences show that H.265 can save 38.83% compared to H.264 at the same PSNR. 2H.265 Real-Time Coding Test In order to test the real-time coding of HEVC, we used the core and 24 core CPU configurations, respectively, and 24 cores, their main frequencies are 3.4GHz, test The code stream is selected at 6m code stream. As shown in Figures 3 and 4. Through the above test, it can be seen that the commercial HEVC encoder can reach the real-time encoding of 30fps under conditions using the 8 core CPU coding. For X265, since the coding speed is slower than the commercial encoder, you need to use more CPUs to reach real-time encoding. Finally, when the CPU of the 24 core is used, the X265 encoder can basically reach the real-time encoding of 25fps. 3.4 Conclusions The H.265 is the next generation of video coding standards in H.264. By the above comparison, the following conclusions can be obtained: H.265 exceeds H.264 in data compression ratios, can replace H.264 to become a wide popular video standard. And H.265 still maintains the advantage of the H.26X series encoder in data compression ratios, which should be said that each step of data compression has been perfectly processed. Based on the above discussion, we believe that the future H.265 will be widely applied. "
Our other product:
