Noise Induced Feature Enhancement and Object Segmentation of Forward Looking SONAR Image

Autonomous Underwater Vehicles (AUVs) are generally relying on Forward Looking SONAR (FLS) data to detect obstacles in front of the vehicle. Presence of obstacle is detected based on the amplitude of the echoed signal sensed by hydrophones. But acoustic signal loses its energy primarily due to transmission loss, absorption and scattering. Due to this, water column image of SONAR is of low contrast and noisy. So an effective preprocessing technique must be employed. This paper is concerned with denoising and contrast enhancement of the FLS image to fortify detection of obstacles. Adaptive dynamic stochastic resonance (SR) has been applied in wavelet domain to enhance the regions of interest (ROI), preceded by the application of Lee filtering to suppress speckle noise. Finally, C-Means clustering based segmentation has been adopted to extract the ROI to calculate the position, size and centre of gravity (CoG) of the obstacle. The proposed algorithm is validated through experimentation carried out on FLS images of Tritech mini king SONAR.