Deep learning-based video coding optimisation of H.265. (4th November 2022)

• Record Type: Journal Article
• Authors: Karthikeyan, C.; Vivek, Tammineedi Venkata Satya; Narayanan, S. Lakshmi; Markkandan, S.; Babu, D. Vijendra; Laddha, Shilpa
• Is Part Of: International journal of engineering systems modelling and simulation. Volume 14:Number 1(2023)
• Page Start: 52
• Page End: 57
• Publication Date: 2022-11-04
• Subjects: deep learning video coding -- DLVC -- high-efficiency video coding -- HEVC/H264 -- rate-distortion -- rate-distortion optimisation -- RDO; Engineering systems -- Computer simulation -- Periodicals; Engineering systems -- Mathematical models -- Periodicals; 620.0042
• Languages: English
• ISSNs: 1755-9758

Abstract

Today's multi-media applications need high video quality with low bitrates. However, it is restricted in its capacity to provide higher quality than earlier coding methods. Deep learning (DL) approaches for video coding have shown compression capacities equal to or better than traditional methods, including high-efficiency video coding (HEVC) methods. The trade-off between compression efficiency and encoding/decoding complexity, optimisation for perceptual nature of semantic dependability, specialisation, and universality, the federalised layout of various deep toolkits, etc. remains unclear. HEVC encoding is more efficient than previous standards. Improved efficiency is driven by intra image prediction, which incorporates more prior directions (35 modes) than previous standards. Its high efficiency comes from balancing encoder complexity and dependability. This article presents DL, which uses a convolutional neural network to predict the best model with the least rate-distortion (RD) and further promotes study into deep learning video coding (DLVC).