M.A. Robertson (USA)
Image Restoration, Wavelet Quantization Noise
Images and motion imagery that are compressed by quan tization of their wavelet coefficients have a reconstruction error equal to the inverse wavelet transform of the coeffi cients' quantization errors. For wavelet compression of im ages, the quantization error in the spatial domain is in gen eral spatially varying and colored. When motion imagery (and three-dimensional imagery in general) is compressed by quantizing the three-dimensional wavelet transform co efficients, the resulting quantization error in the pixel do main is both spatially and temporally varying as well as correlated. This paper presents both experimental and the oretical analysis to demonstrate the nature of the quanti zation error. Equations describing the covariance matrix of the pixel-domain quantization error are provided, which can find future use by algorithms that require accurate noise modeling, such as super resolution and image and video restoration.
Important Links:
Go Back
