A Mosaic Method Based on Feature Matching for Side Scan Sonar Images

摘要

Side scan sonar has been widely used for mapping seabed sediments and structures. The process produces an accurate and continuous sonar image of an area of seabed and is known as mosaicking. In order to obtain high resolution images, sonar transducers are usually towed. However, this method may result in dislocation and distortion of side scan sonar images due to unexpected effects, e.g., waves, and currents. It also causes difficulties for image mosaicking. The present paper proposes an image mosaic method based on feature matching on the seabed. Rough matching and fine matching are introduced in order to achieve a quick approach to really accurate position matching. By preprocessing of side scan sonar images, images of adjacent lanes can be roughly matched by geographic coordinate. During the course of fine matching, the paper proposes the use of the SUSAN algorithm to detect the feature points. Regional mutual information is used for the similarity measurement of feature point matching, and to form a point set of matching features. A transformation matrix is established by the least square method and matching point set. Complete Gaussian pivoting elimination is used to solve equations, which can obtain the final transformation parameters. Images are registered by affine transform. After that, sonar images involve multi-source information in the overlapping area. In order to obtain the final sonar image of the area, wavelet transformation is used to fuse images of the coverage area. After source sonar images are decomposed by wavelet transform, the fusion coefficient of the low-frequency component is determined by the regional energy, while the fusion coefficient of the high-frequency component is determined by the variance of the region. The proposed method can correctly fuse sonar images of adjacent lanes, and this offers a better solution to deal with the dislocation and distortion problems than other traditional methods for image mosaic. The test results show that the proposed method is robust and stable with registration accuracy up to pixel level, thus realizing quite precise automatic registration and fusion, which can be applied for side scan sonar image mosaic.