HEVC 기반 국부 영역 왜곡 인지 확률 모델을 이용한 인지 비디오 부호화

In this paper, a new perceptual video coding (PVC) scheme is proposed based on HEVC, where transform coefficients are suppressed subject to introducing rarely visible distortions in encoding process. In order to apply a perceptual visibility model to flexible coding structures with variable sized DCT kernels of High Efficiency Video Coding (HEVC), we propose a new perceptual visibility model, called the local region distortion detection probability (LP) model. The LP model enables to estimate probability of detecting distortions on any size of DCT block. The experimental results show that the proposed PVC scheme achieves 9.37% bit-rate reduction at the same visual quality compared to the HEVC reference software (HM 11.0).