作者: M.A. Robertson , R.L. Stevenson
DOI: 10.1109/TCSVT.2004.839995(410)
关键词: Image restoration 、 Discrete cosine transform 、 Artificial intelligence 、 Image compression 、 Trellis quantization 、 Compression artifact 、 Transform coding 、 Computer vision 、 Data compression 、 Algorithm 、 Quantization (signal processing) 、 Mathematics
摘要: In lossy image compression schemes utilizing the discrete cosine transform (DCT), quantization of DCT coefficients introduces error in representation and a loss signal information. At high ratios, this introduced produces visually undesirable artifacts that can dramatically lower perceived quality particular image. This paper provides spatial domain model based on statistical noise when quantizing coefficients. The resulting theoretically derived shows general is both correlated spatially varying. some justification to many ad hoc artifact removal filters have been proposed. More importantly, proposed be incorporated post-processing algorithm correctly incorporates correction quantizer error. Experimental results demonstrate effectiveness approach.