End-to-end deep video compression in the feature space has become a key research direction, where effective feature modeling is essential. However, existing methods adopt a single-branch architecture, making it difficult to effectively perform feature modeling on complex video frames. To address this issue, we propose a Context and Fine-grained Fusion Module (CFFM), which adopts a dual-branch architecture for feature modeling. Each branch consists of two Fine-grained Spatial Aggregation Modules and two Dilated Contextual Residual Fusion Modules, focusing on local fine-grained details and global contextual information, respectively. By effectively capturing these complementary features, CFFM significantly enhances the overall feature modeling capability. Additionally, we introduce a Multi-Directional Motion Perception Module (MDMP), which integrates multi-directional and standard convolutions to dynamically expand the receptive field and adaptively model complex motion patterns, improving motion offset estimation. The entire framework is jointly optimized in an end-to-end manner. Experimental results show that our method outperforms traditional codecs and recent deep video compression approaches, achieving a 41.67% bitrate reduction compared to x265 with medium preset under PSNR, and obtaining a 3.54% gain over VVC under the MS-SSIM metric.

Deep learning methods have been achieving good results in the loop filtering in Versatile Video Coding(VVC). Multi-Frame In-Loop Filter of HEVC (MIF) algorithm is one of networks that effectively utilizes the multiple reference frames to enhance the quality of reconstruction. However, it has the disadvantages of low efficiency of the reference frame selection and its quality enhancement network does not fully utilize the inter-frame correlations of video sequence and has large redundancy. To address these problem, an improved multi-reference frame loop filter algorithm based on transformer optimization (T -MIF) is proposed.

The quadtree nested multi-type trees (MTT) partitioning scheme is adopted by the latest generation of video coding standard H.266/VVC. While the performance of the encoder is improved, the complexity has increased by 2.2-5.6 times. In order to reduce the coding complexity, a fast multi-tree partition algorithm based on spatio-temporal information for inter-frame coding is proposed in this paper. The motion characteristics and texture complexity of the current coding block are utilized simultaneously to determine whether the multi-tree partition can be terminated early. For blocks that need to be further divided, the partition direction is determined in advance.

To address the problem of large computational redundancy in the geometric prediction merge mode with motion vector refinement (GPM with MMVD), a new decision algorithm based on CU gradient is proposed. By comparing the mean value of the gradient in four directions to determine whether GPM can be terminated early. The advance decision of GPM partition mode can be determined by the calculated gradient direction of CU. Meanwhile, the proposed algorithm saves coding time by reducing the number of candidate modes during the mode selection process.

360-degree video has been developed rapidly and widely used in recent years.360Lib is a test platform (software) for JVET panoramic video, hosted by Subversion code. The 360Lib in this article is integrated into the HM.At this stage, coding 360 degrees video needs to convert video according to ERP, EAP, CMP and other formats, then encode with HEVC, and then reverse transform to 360 degree video.Since 360lib just transforms the projection format, it still uses HEVC to code, and does not use the 360-degree video's characteristics. Based on the characteristics of 360lib, this paper proposes a A CU Split Early Termination Algorithm. Experimental results show that the proposed fast algorithm provides an average time reduction rate of 32% compared to the reference HM-16.16+360lib4.0, with only 1.3% BD-rate increase.

Virtual Reality 360 degree video is spherical video. It allows the viewer to freely choose the field of view from any angle, and the video resolution must be high enough for every field of view. During coding process, spherical video is projected into 2D plane video and it has ultra-high resolution and larger data compared to traditional video. Considering reducing the high spatial redundancy of virtual reality 360 degree video, and reducing the computational complexity of intra coding, while ensuring the integrity of the details of each field of view, a fast intra prediction algorithm for virtual reality 360 degree video based on improved Rough Mode Decision (RMD) is proposed. The experimental results show that the proposed algorithm brings 38.73% time saving with only 1.4% Bjontegaard delta rate increase in All-Intra mode.

Virtual reality technology has become increasingly popular in recent years. It has more and more attempts and applications in online teaching, traffic safety, video and so on. Virtual reality video is popular because of its immersive feel. 360-degree video is one of the virtual reality videos. This type of video has a very high resolution, which makes its encoding time considerably long. To reduce the encoding complexity of virtual reality 360 video, this study proposed a fast early termination of CU partition and mode selection algorithm in HEVC. The proposed algorithm determines whether to terminate CU early or directly further divide in the CU partition process. On the other hand, the proposed algorithm saves time by reducing the number of candidate modes during the mode selection process. Experimental results show that the proposed algorithm can reduce the coding time by 54% and the BD-rate loss is only 1.4%.

As virtual reality becomes more popular, 360-degree video coding becomes challenging. Projected videos of 360-degree videos and traditional videos are both planar videos, but the projected videos have distortion whose degree depends on the latitude. Traditional coding algorithms cannot effectively adapt to this feature, and 360-degree videos typically have high resolution and frame rate, which results in a high coding complexity. In this study, a fast prediction unit (PU) early termination algorithm based on weighted mean square error (WMSE) of 360-degree video is proposed. In the proposed algorithm, WMSE is used as a basis for the early termination of further PU partitioning. First, the full intra prediction process of the current CU is performed. After that, the similarity between the current CU and its four sub-CUs is calculated using WMSE for 2N×2N; and distortion between predicted and original blocks is calculated using WMSE for N×N. the similarity and distortion is used to terminate PU partitioning. The experimental results show that the algorithm achieves a 31% time reduction, and an average of only 0.3% of the luma Bjontegaard delta rate (BD rate) increases.

High Efficiency Video Coding (HEVC) introduces 35 intra prediction modes and a flexible quad-tree coding structure, which remarkably improve the coding efficiency. However, in intra prediction, the cost computation in Rough Mode Decision (RMD) process and Rate Distortion Optimized (RDO) process suffers from a pretty high complexity compared with H.264. To deal with this problem, a modified RMD process is proposed, in which all 35 prediction modes are divided into groups according to their phase angle to reduce the candidate modes.

High-Efficiency Video Coding (HEVC) is the newest video coding standard and presents improved efficiency. However, the coding computational complexity of this process is significant because of its coding structure called quad-tree structure coding units (CUs) for improving compression efficiency. In this paper, a fast algorithm based on image complexity of current CU is proposed to reduce the computational complexity caused by CU partition. Using the number of colors, entropy, and contrast can distinguish the image complexity of different CUs. Such complexity is related to whether the CU needs to be partitioned or its rate distortion optimization can be skipped. Coding bits of current CU are used as qualifications to determine whether to partition the current CU. Experimental results show that the proposed algorithm can reduce coding time by 35% on average and exhibits an acceptable loss of Bjøntegaard delta bit rate of 0.8% compared with the default algorithm in HM-16.11 for encoding natural sequences.

Summary form only given. Screen content has been an important extension of high efficiency video coding widely used in wireless display and screen sharing. Reconstructed screen content images have distortions because of the quantization of transforming coefficients. Hence, this paper proposes an enhanced block-matching and 3D filter (BM3D) to attenuate the distortion by block classification and block segmentation. Experimental results show that the proposed algorithm achieves not only a peak signal-to-noise ratio increase but also subjective visual quality improvement for HEVC screen content images denoising.

To satisfy the rapidly increasing demands for high resolution video, the Joint Collaborative Team on Video Coding (JCT-VC) developed a new video coding standard, i.e. High-efficiency Video Coding (HEVC). With the same perceptual video quality as H.264, HEVC can achieve a bit rate reduction of approximately 50%. However, the quad tree structured coding units (CUs) cause high computational complexity. In this paper, a novel algorithm is proposed to decrease the computational complexity of HEVC intra coding. We mainly use the total coding bits of CUs as threshold to terminate CUs' partition. Furthermore, we also consider the spatial correlation between the current LCU and its neighbor LCUs to predict the current largest CU minimum depth so that we can skip some certain CUs' rate distortion cost checking.