Low-Complexity Fast CU Classification Decision Method Based on LGBM Classifier

At present, the latest video coding standard is Versatile Video Coding (VVC). Although efficiency of VVC significantly improved compared to previous generation, High-Efficiency (HEVC), it also leads a sharp increase in complexity. improves HEVC by adopting quadtree with nested multi-type tree (QTMT) partition structure, which has been proven be very effective. This paper proposes low-complexity fast unit (CU) decision method based on light gradient boosting machine (LGBM) classifier. Representative features were extracted train classifier matching framework. Secondly, new CU framework was designed for VVC, could predict advance whether divided, divided (QT), and horizontally or vertically. To solve multi-classification problem, technique creating multiple binary classification problems used. Subsequently, multi-threshold decision-making scheme consisting four threshold points proposed, achieved good balance between time savings efficiency. According experimental results, our significant reduction encoding time, ranging from 47.93% 54.27%, but only Bjøntegaard delta bit-rate (BDBR) 1.07%~1.57%. Our showed performance terms both