Preprocessing and Detection

Wavelet denoising reduces high frequency noise.

Noise level at each wavelet scale is estimated separately.

This defines a threshold for zeroing wavelet coefficients.

Other wavelet coefficients are shrinked according to local variance estimation.

After inverse wavelet transform, the image is renormalized.

A matched filter is used in the detection phase.

Its mask is optimized to fit the mine profile at three target distances.

'Hot spots' are grouped based on the intensity level of the filter convolution.

Detection takes place using various decision rules that are applied to those suspected areas.