Article Preview
TopIntroduction
With the rapid development of video applications for video resolution and storage (Peng and Cohen et al., 2020), High-Efficiency Video Coding (HEVC/H.265) will not be able to meet the needs of today's diverse demand (Mahdavi & Hamzaoglu, 2021; Yang et al., 2021; Jiang et al., 2020). In order to improve video coding technology, the Joint Video Exploration Team (JVET) developed Versatile Video Coding (VVC) (Bross et al., 2021; Acharjee & Chaudhuri et al., 2022) based on HEVC, which was finalized as a new video compression standard in July 2020 with the coding technologies and methods of HEVC retained and the new encoding tools introduced, such as Binary Tree (BT) and Trident Tree (TT) in a Multi Type Tree (MTT) structure (Chen et al., 2018), and Multiple Transform Selection (MTS). Although the coding performance of VVC has been further improved (Chen & Ye et al., 2020), the coding tools adopted also bring huge complexity to the coding process (Yang et al., 2020). Therefore, reducing coding complexity without losing coding quality has become a key research focus in recent years.
Scholars have done a lot of algorithm research to reduce the computation complexity of intraframe coding and have made certain achievements, which can be summarized into two categories. The first category of methods speeds up the process of Coding Unit (CU) partitioning based on the correlation features of texture, depth, and adjacent blocks. For example, a fast intraframe encoding algorithm was proposed by skipping redundant coding block structure and unnecessary direction partition in Zhang et al. (2022). A fast multitype-tree split decision algorithm based on the similarity between adjacent subregions in horizontal and vertical directions was proposed in Liu et al. (2021). A new intra-VVC frame-based fast CU-splitting algorithm based on cross-block gradient information acquired from CU decomposition was proposed in Liu et al. (2021). The second category of methods has been another research hotspot, including algorithms based on Convolutional Neural Networks (CNN) or Machine Learning (ML). For example, in Wu et al. (2021), a fast support vector machine CU splitting algorithm was proposed that predicts the splitting mode of CU based on texture information and preemptively terminates redundant splitting. Zhou et al. (2021) proposed a deep CNN based on texture classification to predict the division of a Coding Tree Unit (CTU) and trained a CNN structure to determine whether the coding unit should be terminated in advance. Saldanha et al. (2022) proposed a partitioning decision based on a Light Gradient Boosting Machine (LGBM) classifier, which uses effective features extracted from texture, encoding, and contextual information, as well as powerful machine learning models. In He et al. (2021), a Random Forest (RF) algorithm was proposed to reduce the complexity of the CU partition. The algorithm trains an RF classifier based on the texture complexity between adjacent CUs to predict whether the CU partition process terminates early to accelerate the partition process of VVC frames.