Reverberation reduction in ultrasonic images via predictive deconvolution

Reverberations are usually observed in ultrasonic nondestructive testing (NDT) results and medical images. They can blur the echoes from the targets, as a result, lack the accuracy of material examinations as well as clinical diagnoses. The reverberation can be recognized or suppressed via some known techniques. However, the repetitive period and reverberation waveform may vary and differ from their primary reflections. Under such situation, the known techniques will not work efficiently.In order to improve these situations, a predictive deconvolution technique is proposed to suppress reverberations in ultrasonic images. Because the appropriate prediction lag (α) and filter length (N) in the predictive deconvolution can be specified, the reverberations with specified periodicities between α and α+N−1 are able to be suppressed, and the length of the desired output wavelet can be controlled as α. In addition, a better output of predictive deconvolution can be obtained if (1) the prediction lag is slightly less than the repetitive period of the reverberation, and (2) the prediction filter length is greater than the length of the source echo. Experiments show that reverberations with different repetitive periods in B-scans are greatly reduced via predictive deconvolution, though the waveform of the source echo changes slightly when propagates in the specimen. Once buried in reverberations, now the images of the holes located at different depths, with different sizes, can be easily observed.