Wavelet Analysis of Print Defects

This paper examines wavelet analyses for detecting and characterizing printing defects. Positive wavelet features for this application include localized space-frequency properties for characterizing defects of limited spatial support (in contrast with the Fourier analysis, which has no spatial localization over the analysis region). In addition, the scale based analyses of wavelets mimic properties of human vision system (HVS) that can be useful for developing thresholds consistent with visual masking and sensitivity. This study examines simulated defects in monochromatic images. Examples illustrate how defects, such as banding, graininess, and streaking appear in the wavelet domain. Wavelet statistics for characterizing printing defects are suggested and their performance tested through Monte Carlo simulations. Simulation results compare the performance of several popular wavelet kernels, such as Daubechies, symlets, and biorthogonal splines. The influence of wavelet properties, such as smoothness and symmetry, on performance is discussed. Detection and estimation results of defects in noise show that symlets generally perform well for characterizing all defects considered. While all wavelets performed well for the banding defects, the biorthogonal spline wavelets performed significantly worse for estimating graininess defect properties.