Perceptual Quality Enhancement for Compressed Video With High-Frequency Details and High-Dimensional Features

The ultimate goal of video compression enhancement is to improve the perceptual quality of compressed videos, which is highly dependent on high-frequency details and high-dimensional features. However, most existing approaches focus on enhancing the objective quality of compressed videos, which may not necessarily correspond to the perceptual quality perceived by humans. This paper proposes a generative adversarial network based approach called HFHD-GAN to improve the perceptual quality of compressed videos. The HFHD-GAN approach consists of three main modules: the multiscale directional convolutional network (MDCN) is designed to extract high-frequency details adaptively; the directional feature attention (DFA) is to reduce the high-frequency noise and the multilevel directional feature discriminator (MDFD) is for recovering high-dimensional features of videos. Experimental results show that the proposed method achieves outperformance of perceptual quality with sharper edges and vivid details, compared to state-of-the-art methods for compressed video quality enhancement. This approach has potential applications in video streaming, conferencing, and surveillance.