Fuzzy Based Adaptive Deblocking Filters at Low-Bitrate HEVC Videos for Communication Networks

• Published in: Computers, materials & continua materials & continua, 2021, Vol.66 (3), p.3045-3063
• Main Authors: Gandam, Anudeep, Singh Sidhu, Jagroop
• Format: Article
• Language: English
• Subjects: Adaptive filters; Adaptive systems; Blocking; Boundaries; Communication networks; Communications networks; Complexity; Experimentation; Multimedia; Outliers (statistics); Parallel processing; Post-processing; Quadrants; Video compression
• ISSN: 1546-2226; 1546-2218; 1546-2226
• DOI: 10.32604/cmc.2021.013663

In-loop filtering significantly helps detect and remove blocking artifacts across block boundaries in low bitrate coded High Efficiency Video Coding (HEVC) frames and improves its subjective visual quality in multimedia services over communication networks. However, on faster processing of the complex videos at a low bitrate, some visible artifacts considerably degrade the picture quality. In this paper, we proposed a four-step fuzzy based adaptive deblocking filter selection technique. The proposed method removes the quantization noise, blocking artifacts and corner outliers efficiently for HEVC coded videos even at low bit-rate. We have considered Y (luma), U (chroma-blue), and V (chroma-red) components parallelly. Finally, we have developed a fuzzy system to detect blocking artifacts and use adaptive filters as per requirement in all four quadrants, namely up 45°, down 45°, up 135°, and down 135° across horizontal and vertical block boundaries. In this context, experimentation is done on a wide variety of videos. An objective and subjective analysis is carried out with MATLAB software and Human Visual System (HVS). The proposed method substantially outperforms existing post-processing deblocking techniques in terms of YPSNR and BD_rate. In the proposed method, we achieved 0.32–0.97 dB values of YPSNR. Our method achieved a BD_rate of +1.69% for the luma component, −0.18% (U) and −1.99% (V) for chroma components, respectively, with respect to the state-of-the-art methods. The proposed method proves to have low computational complexity and has better parallel processing, hence suitable for a real-time system in the near future.