Based On The Superpixels With Low-rank Representation Method Of Image Segmentation

As the marine resource exploit has been increasingly serious today,the sonar image with high resolution has great significance to explore ocean.However,the image segmentation technique is not only an pivotal Step for image analysis but also the premise for target identification.Therefore the paper mainly study a neoteric image segmentation technique----The paper can be divided to 4 steps:Step 1: Segment the image on a superpixels levels.Therefore,two image segmentation techniques which are appropriate for grayscale images are chosen.Finally,the paper also choose the Turbopixels method which is more befitting for the sonar image as the first step.Step 2: Compare and analyse the features.The paper processes diversity comparison in the aspects of gray level co-occurrence matrix and gradient-gray level co-occurrence matrix,then pick up the features which have obvious distinction and process the feature extraction,and generate the eigenmatrix.Step 3: Process a low-rank to the eigenmatrix and the address can express Robust noise?Improve the similarity measure.By solving the Conditional extreme value formula by the method of EALM,we can get the optimal similarity matrix.Step 4: Generate the segmentation result which is shaped like a map by processing an N-CUT to the optimal similarity matrix.In fact,the last to steps use the idea of Subspace Segmentation.However,the final segment image is the result of the cluster to the superpixels.Because of the superpixels Segmentation ways has a feature of segmenting the image with a distinct and accurate edge,the segmenting result coincide the original image perfect.