Ambient noise imaging; enhanced spatial correlation algorithms and a way to combine independent images for improved stability and false-alarm rejection

Acoustic daylight, second-order temporal and second-order spatial imaging are three algorithms that have been devised for ambient noise imaging. This paper proposes two enhancements (enhancement I and II) to improve the image quality obtained by spatial imaging. Enhancement I repeats the original spatial imaging algorithm for a limited number of times and achieves an improvement of up to 10.5 dB. Enhancement II uses linear programming to maximize the separation between the target and non-target sets, achieving up to 2 dB benefit. Enhancements I and II are combined to produce a better result than enhancement I alone with its highest improvement at a further 7.5 dB. We also propose a fusing algorithm, using K-mean clustering with a validity measure, to combine images produced by different algorithms so as to improve the stability and false alarm rejection of the images. An average additional improvement of 2 dB is obtained when compared to direct fusing without using clustering. Better results are usually obtained if the features (i.e. results from different algorithms at a particular pixel) at one channel are sorted prior to clustering.