A Multi-Scale Denoising Method for Forward-Looking Sonar Image

Underwater object detection and recognition rely heavily on the quality of sonar images, which can be significantly impacted by noise. This paper proposes a multi-scale denoising method for forward-looking sonar images. The proposed method firstly decomposes sonar image into three layers containing edge, texture and content information by the Laplacian pyramid model. Next, the adaptive median filter is applied to the edge layer while the modified adaptive local filter is applied to the texture layer and the content layer is completely retained. Finally, all processed layers are synthesized by Laplacian pyramid inverse transform. Experimental and comparative results on real and simulated sonar images demonstrate that the proposed method effectively reduces speckle noise and improves image quality.