Spectral Ringing Artifacts in Hyperspectral Image Data Compression
M. Klimesh, Aaron B. Kiely, Hua Xie, Nazeeh Aranki · Kluwer Academic Publishers eBooks · 2006
When using 3-D wavelet transforms for hyperspectral image compression, systematic variations in signal level of different spectral bands can cause widely-varying mean values in spatial planes of spatially low-pass subbands. Failing to account for this phenomenon can have detrimental effects on image compression, including reduced effectiveness in compressing spatially low-pass subband data, and biases in some reconstructed spectral bands. These effects can be mitigated by subtracting the mean value from each spatial plane of each spatially low-pass subband, or by modifying the wavelet decomposition to perform extra stages of spatial decomposition in spatially low-pass subbands, or by a combination of these approaches. We presented examples illustrating that these methods offer similar improvements in rate-distortion performance. Both approaches reduce biases in reconstructed spectral bands and provide an improvement in subjective reconstructed image quality. The modified decomposition has the advantage that it does not have a tendency to produce visible boundaries between error-containment segments.