QUALITY ASSESSMENT FOR DEEP LEARNING BASED VIDEO CODING
Project no. 22956 · Status: Kapanmış (Yayın bekleniyor) · 2021
Summary
<p>Compression methods used to effectively store and transmit video content in the digital environment have become even more important with the increasing prevalence of high-resolution videos today. In this regard, in addition to traditional codecs, deep learning-based new generation methods also attract attention. In this study, the H.265 (HEVC) codec and a deep learning-based codec called LHBDC (End-to-End Rate-Distortion Optimized Learned Hierarchical Bi-Directional Video Compression) were compared on 1080p videos encoded at certain bitrate values. During the comparison process, both objective metrics and subjective evaluations were used. By combining the results of objective metrics with the subjective perceptions of users, the advantages and disadvantages of the deep learning-based approach compared to the traditional method were analyzed. The findings showed that LHBDC has the potential to deliver higher quality, especially in high-motion scenes or low bit rates. This study makes an important contribution to evaluating the performance of different coding solutions with more realistic methods and to respond to next generation video transmission/storage needs.</p>