Mode skipping for screen content coding based on Neural Network Classifier

作     者：Elsawy, Nabila Sayed, Mohammed S. Farag, Fathi 

作者机构：Zagazig Univ Dept Elect & Commun Engn Zagazig Egypt Egypt Japan Univ Sci & Technol Dept Elect & Commun Engn New Borg El Arab City Egypt 

出 版 物：《JOURNAL OF REAL-TIME IMAGE PROCESSING》 (实时图像处理杂志)

年 卷 期：2021年第18卷第6期

页      面：2453-2468页

核心收录：

学科分类：0808[工学-电气工程] 1002[医学-临床医学] 08[工学] 0812[工学-计算机科学与技术（可授工学、理学学位）] 

主　　题：HEVC Screen content coding Neural network Intra-prediction 

摘      要：The Screen Content Coding Extension in High-Efficiency Video Coding standard (HEVC-SCC) promotes the capabilities of HEVC in coding screen content videos (SCVs) using new techniques, which improves coding efficiency dramatically. These new techniques depend on the distinguished features of SCV such as repeated patterns, limited number of colors, sharp edges, and non-noisy regions. Nonetheless, this coding efficiency comes at the cost of enormous computational complexity. In this paper, a new technique is proposed to save encoding time while conserving coding efficiency. The proposed algorithm selects the suitable mode for each Coding Unit (CU) and skips unhelpful modes by two methods. Two methods depend on skipping unwanted modes by Neural Network Classifiers. The first classifier is Neural Network Classifier Based on Current Depth Features (NNC_CF), which depends on the CU current depth features. The second one is Neural Network Classifier Based on Parent Depth Features (NNC_PF);the Parent depth features are considered the input of this classifier. The simulation results demonstrate the efficacy of the proposed scheme.