Super-Resolution Reconstruction of Compressed Video Based on Noise Distribution Property
基于噪声分布特性的压缩视频超分辨率重建

This paper models the process of video compression, DCT quantization noise and motion estimation noise with exploiting the quantization step size and motion information embedded in the bit-stream. Together with the additive noise term of imaging, the proposed total noise term adaptively adjusts for different quantizers. With a Huber-Markov Random Field (HMRF) as the prior model, the gradient descent algorithm and MAP super-resolution reconstruction are presented and their performances are also analyzed. Simulation results show that proposed algorithm obtains better objective and subjective performances.