Image Denoising via Trained Dictionaries for the Time-frequency Image of Underwater Acoustical Plus Signals

Detecting the tracks of spectrograms is an important step of estimating the instantaneous frequency of underwater acoustical plus signals. Image processing applied to this area treat the spectrogram as an image containing features to be extracted. And the difficulties lie in that the effect of image processing is disturbed by channel disturbance and the strong ambient sea noise. This paper presents an adaptive method of image denoising which applies to the time-frequency image of underwater acoustical plus signals. The approach taken is based on sparse and redundant representations over trained dictionaries. Using the prior information of ambient sea noise model, a similarity constraint term is introduced to the improved K-SVD algorithm, we can get a new objective function. Experiments on the time-frequency image of underwater acoustical plus signals demonstrate the effectiveness of the proposed algorithm.