Stationary Wavelet Transform for denoising Pulsed Thermography data: optimization
of wavelet parameters for enhancing defects detection
Abstract
Innovative denoising techniques based on Stationary Wavelet Transform (SWT) have started being applied to
Pulsed Thermography (PT) sequences, showing marked potentialities in improving defect detection. In this contribution, a
SWT-based denoising procedure is performed on high and low resolution PT sequences. Samples under test are two
composite panels with known defects. The denoising procedure undergoes an optimization step. An innovative criterion
for selecting the optimal decomposition level in multi-scale SWT-based denoising is proposed. The approach is based on
a comparison, in the wavelet domain, of the information content in the thermal image with noise propagated. The optimal
wavelet basis is selected according to two performance indexes, respectively based on the probability distribution of the
information content of the denoised frame, and on the Energy-to-Shannon Entropy ratio. After the optimization step,
denoising is applied on the whole thermal sequence. The approximation coefficients at the optimal level are moved to the
frequency domain, then low-pass filtered. Linear Minimum Mean Square Error (LMMSE) is applied to detail coefficients at
the optimal level. Finally, Pulsed Phase Thermography (PPT) is performed. The performance of the optimized denoising
method in improving the defect detection capability respect to the non-denoised case is quantified using the Contrast Noise
Ratio (CNR) criterion.
download