An improved spectral clustering sonar image segmentation method

To solve the problem that many image segmentation methods cannot be applied to forward looking sonar image accurately, an improved level set segmentation method was proposed in this paper. Firstly, the level set evolution without re-initialization was introduced. Secondly the different characteristics of forward looking sonar image from the optical image were analyzed, and we got the factors affecting segmentation. Then, to overcome these negative effects, this paper did preprocessing by morphological top-hat and bottom-hat transformation, and carried on level set method without re-initialization to construct an improved level set sonar image segmentation system. Finally, our method was compared with the traditional level set method in computer experiments. Simulation results show that it is more adapted to forward looking sonar image segmentation.

Image segmentation method combined with Markov random field (MRF) model and regional growth image in this paper is proposed to achieve the accuracy of sonar image segmentation. First of all, the characteristics of field and the label field of sonar image is established through the MRF model; secondly, the further image segmentation is processed by region growing based on the initial image segmentation by ICM algorithm, and this reduced segmentation plaques caused by the MRF method on image segmentation. Finally, experimental results show that the method in this paper can obtain the image segmentation effectively.

The mosaic of sonar images is more difficult than the mosaic of traditional optical images due to their poor quality and the difficulty in extracting feature points. The existing mosaic methods of sonar images have a series of problems, such as low correct matching rate, large cumulative errors and high requirements for the quality of collected sonar images. In this paper, we proposed a high precision method to implement the mosaic of the underwater sonar video image sequence. Firstly, Accelerated Unsharp Masking (AUSM) algorithm is proposed to preprocess the original image. Then we extract KAZE feature points from preprocessed sonar images. A matching method combining multiple matching strategies with Progressive Sample Consensus (PROSAC) algorithm is followed to complete the image registration. Weighted fusion method and a region of interest (ROI) acquisition method based on the slope of right border is used to optimize the mosaicked image. Finally, we can obtain a high-quality panoramic image of underwater sonar video by a global mosaic strategy. Mosaic experiments on the sonar video image sequence collected by multi-beam sonar demonstrate that the proposed method in this paper can increase the number of feature points by about 70% and make the correct matching rate higher than 70%. The proposed method also has good robustness and the cumulative error during multi-image mosaic is less.

This paper proposes a fast approach to spectral image segmentation. In the algorithm, two popular techniques are extended and applied to spectral color images: the mean-shift filtering and the kernel-based clustering. We claim that segmentation should be completed under illuminant F11 rather than directly using the original spectral reflectance, because such illumination can reduce data variability and expedite the following filtering. The modes obtained in the mean-shift filtering represent the local features of spectral images, and will be applied to segmentation in place of pixels. Since the modes are generally small in number, the eigendecomposition of kernel matrices, the crucial step in the kernel-based clustering, becomes much easier. The combination of these two techniques can efficiently enhance the performance of segmentation. Experiments show that the proposed segmentation method is feasible and very promising for spectral color images. Keywords: spectral color image, illuminant F11, mean-shift filtering, spectrum extension, kernel-based clustering

The spectral clustering algorithm's space complexity is O(n <sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">2</sup> ), while time complexity is O(n <sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">3</sup> ). When dealing with large amounts of data, the memory will overflow and run-time is too long. For the general problem of spectral clustering, if the clustering data of sub-problem between the original problem has the same probability distribution, it can be applied to divide and conquer strategy for the problem of spectral clustering, by the spectral clustering results of sub-problems to get the spectral clustering results of original problem. To spectral clustering image segmentation as a research object, we will discuss the divide and conquer strategy for spectral clustering in this paper. Experiments show that the application of divide and conquer method for spectral clustering image segmentation, we can get a perfect performance in image segmentation.

In this work, we propose a new algorithm for spectral color image segmentation based on the use of a kernel matrix. A cost function for spectral kernel clustering is introduced to measure the correlation between clusters. An efficient multiscale method is presented for accelerating spectral color image segmentation. The multiscale strategy uses the lattice geometry of images to construct an image pyramid whose hierarchy provides a framework for rapidly estimating eigenvectors of normalized kernel matrices. To prevent the boundaries from deteriorating, the image size on the top level of the pyramid is generally required to be around 75 x 75, where the eigenvectors of normalized kernel matrices would be approximately solved by the Nyström method. Within this hierarchical structure, the coarse solution is increasingly propagated to finer levels and is refined using subspace iteration. In addition, to make full use of the abundant color information contained in spectral color images, we propose using spectrum extension to incorporate the geometric features of spectra into similarity measures. Experimental results have shown that the proposed method can perform significantly well in spectral color image segmentation as well as speed up the approximation of the eigenvectors of normalized kernel matrices.

Aiming at the problems such as great noise disturbance and poor performance of traditional arithmetic in the field of underwater sonar image segmentation, based on analyzing grey-level distribution of sonar image and the Markov random fields theory (MRF), several groups of gray-level distribution model for shadow area and background area were introduced. The adaptive maximum-entropy arithmetic was introduced to initialize the segmentation, and the image pre-segmentation arithmetic based on fractal theory was presented. Performance of these models was simulated and analyzed. Simulation results indicate that the method given by this paper has good precision and great robustness, furthermore this method has little dependence on image pre-processing. The analyzed conclusions have guiding meaning for understanding and analysis of sonar image.

Vascular segmentation plays an important role in medical image analysis. A novel technique for the automaticextraction of vascular trees from 2D medical images is presented, which combines Hessian-based multiscale filtering and a modified level set method. In the proposed algorithm, the morphological top-hat transformation is firstly adopted to attenuate background. Then Hessian-based multiscale filtering is used to enhance vascular structures by combining Hessian matrix with Gaussian convolution to tune the filtering response to the specific scales. Because Gaussian convolution tends to blur vessel boundaries, which makes scale selection inaccurate, an improved level set method is finally proposed to extract vascular structures by introducing an external constrained term related to the standard deviation of Gaussian function into the traditional level set. Our approach was tested on synthetic images with vascular-like structures and 2D slices extracted from real 3D abdomen magnetic resonance angiography (MRA) images along the coronal plane. The segmentation rates for synthetic images are above 95%. The results for MRA images demonstrate that the proposed method can extract most of the vascular structures successfully and accurately in visualization. Therefore, the proposed method is effective for the vascular tree extraction in medical images.

In recent years, there has been a notable surge in the development of methods dedicated to classifying hyper spectral image samples. However, the inherent high dimensionality of hyper spectral data presents significant challenges for traditional supervised classification algorithms. Prior studies have attempted to address these challenges by incorporating active learning methods for hyper spectral data segmentation. Nevertheless, a crucial drawback of active learning lies in its dependence on labeled data samples, leading to less improved classification results for unlabeled data samples. In response to this limitation, our work introduces Supervised Active Learning (SAL) techniques that harness both spectral and spatial information. The key innovation of our research lies in the introduction of a novel hyper spectral image segmentation approach based on the k-dependence Bayesian (KDB) model. We estimate the maximizer of the posterior marginal using Symmetric Iterative Proportional Fitting (SIPF) within this KDB model. By implementing the KDB model for hyper spectral image segmentation, our goal is to enhance the classification accuracy of our SAL classifiers. The SAL framework integrates shape prior probability estimation, class density results from Improved Multi-Level Logistic (IMLL) methods, and the maximization of the posterior marginal estimated through Symmetric Iterative Proportional Fitting (SIPF). We evaluate the accuracy of our proposed MPM-SIPF-SAL system using the real ROSIS Hyper Spectral Dataset, demonstrating that our system outperforms existing methods in terms of classification accuracy.

This paper proposes a method to detect underwater objects using sonar image simulator and convolutional neural network (CNN). Instead of simulating very realistic sonar images which is computationally complex, we implemented a simple sonar simulator that calculates only semantic information. Then, we generated training images of target objects by adding randomized degradation effects to the simulated images. The CNN trained with these generated images is robust to the degradation effects inherent in sonar images and thus can detect target objects in real sonar images. We verified the proposed method using the sonar images captured at sea through field experiments. The proposed method can implement object detection more easily because it only uses simulated images instead of real sonar images which are challenging to acquire. The proposed method can also be applied to other sonar-image-based algorithms.

This article handles the possibility of using classification and segmentation of sidescan sonar images for long term registration. In our case, long term registration means to find the displacement between two images which can have been mapped with many weeks or many months between them. The aim of this study is to help AUVs (autonomous underwater vehicles) to navigate, in particular to correct the drift of navigation sensors. This type of positioning raises two sorts of problems, which come from image properties: spatial variability and temporal variability. The first one is caused above all by the sonar geometry and appears for example like a modification of the shape or the position of the shadow according to the point of view. This effect can also be seen in textures, for example on megaripples of sand, which can be more or less visible depending on the point of view of the sonar. The second one is more the consequence of the seafloor physics: between two images, mapped at different times, some elements may have changed. An obvious example is the presence of evanescent objects like fishes but this variability can also be seen on sediments, which borders can move due to local bottom dynamics. With the aim to solve these problems and to provide reliable landmarks for matching, we decided to classify and segment the images. The data have first been corrected from TVG (time varying gain) effects and despeckelised in order to process images which are more representative of the seafloor. The basis idea is to use a supervised method of classification. To do that, we consider some parameters which are coming from a decomposition by Gabor filters, in order to segment with linear discriminant analysis and use of the nearest neighbour method. Registration needs accurately localised landmarks: so, this operation is split in several stages, refining step by step the classification, in order to obtain a map which describes the seafloor with the most possible detailed frontiers. Then, we present the obtained results considering five texture classes: rocks, megaripples, sand, mud and shadow. These several areas and their frontiers are the basis landmarks to match the images. However, before using the segmentation, we must check its reliability. So, it appears that the frontiers, though they are realistic, are not accurate enough to make a registration precise to few pixels, especially in rock areas. Similarly, according to the orientation of the ripples, they may be seen as ripples or sand. These observations are due to the directivity of the sonar, which caused these effects on the segmentation. To do registration, we must take into account these problems. So, the results of the segmentation will be used only for a coarse registration, in order to find quickly the area of interest but also to assess the reliability of our registration (matching on ripples areas is a priori less reliable than on rocks areas). The results of registration are shown, proving the good adequacy between reference image and test image. Others methods, more quantitative, will be able to be tested, to refine the results.

Marine sonar image noise absorption under water can result in poor image quality, low image resolution and blurred target contour. This paper proposes a systematic framework including underwater image fuzzy preprocessing method, and clustering based on the principal component analysis (PCA) of underwater targets. Underwater images were denoised adopting fast median filtering algorithm with mean acceleration. The fuzzing mathematics were not only developed with the aim of improving the visual quality of underwater sonar images, but also involved in the sonar image segmentation for extracting the target area from the suspicious area. Both the geometric and statistical features were treated as features in PCA algorithm. As demonstrated by the experimental results, visual quality of the sonar image was improved, multi-target threshold segmentation was achieved and multiple targets could be analyzed and clustered for stably tracking AUV.

A wavelet transform-based image segmentation method

Diver target automatic detection is indispensable for underwater defense systems, such as the unmanned harbor surveillance system. It is a very challenging task due to various poses and intensity features of diver target. In addition, the background noise in sonar images is complex, which also makes the task more difficult. In this paper, we propose a diver detection method based on saliency detection for sonar images. On the basis of studying the characteristics of diver sonar images, we first decompose the original sonar image and perform median filtering on it, which can significantly improve the quality of the sonar image saliency map. We employ saliency detection technique based on frequency analysis to segment the acoustic highlight region from its surroundings. This segmentation region roughly locates the diver target and generates a region of interest (ROI). We then extract the acoustic shadow region in ROI, which contributes to furtherly improve the localization accuracy. Finally, we merge the segmented highlight region and the extracted acoustic shadow region and compute the minimum outer rectangle of the merged region. Experimental results validate that the proposed method can well detect and locate the diver target, and it can also satisfy the demands of real-time application, and there is almost no false alarm in this method.

This work deals with unsupervised sonar image segmentation. We present a new estimation segmentation procedure using the recent iterative method of estimation called Iterative Conditional Estimation (ICE) [1]. This method takes into account the variety of the laws in the distribution mixture of a sonar image and the estimation of the parameters of the label field (modeled by a Markov Random Field (MRF)). For the estimation step, we use a maximum likelihood technique to estimate the noise model parameters, and the least squares method proposed by Derin et al. [2] to estimate the MRF prior model. Then, in order to obtain an accurate segmentation map and to speed up the convergence rate, we use a multigrid strategy exploiting the previously estimated parameters. This technique has been successfully applied to real sonar images 1, and is compatible with an automatic processing of massive amounts of data.