Region Adjacency Graph Research Articles

Adaptive superpixel based Markov random field model for unsupervised change detection using remotely sensed images

ABSTRACTThis study presents an adaptive superpixel based Markov Random Field (ASP_MRF) model for unsupervised remotely sensed images change detection. Firstly, the difference image is generated by change vector analysis (CVA) and the zero parameter version of the ‘simple linear iterative clustering’ method (SLICO) is applied on the difference image to obtain the superpixel map. Then, the superpixel map is initially labeled as changed and unchanged class by Fuzzy c-means (FCM) clustering method. Thirdly, the region adjacent graph (RAG) is built on the superpixel map to model the spatial constraints between the adjacent superpixels. Specially, the spectral dissimilarity between the adjacent superpixels and the label fuzziness of the neighbored superpixels were incorporated in the RAG. Lastly, The initial labels of the superpixel map are iteratively refined with ASP_MRF to generate the final change map. The experimental results prove that ASP_MRF obtained the most accurate change map and outperformed the results by pixel level MRF and superpixel based MRF, which verifies the effectiveness of ASP_MRF.

Research on image segmentation method using a structure-preserving region model-based MRF

This paper proposes a structure-preserving region model for machine images. Under the Bayesian framework, the proposed model is combined with MRF (Markov random field) to offer a new method for the segmentation of machine images. The structure-preserving region model aims to deal with problems with MRF-based segmentation on parameter estimation and optimization. Construction of the structure-preserving region model involves two processes. The bilateral filter algorithm is first applied to machine images to remove noise and restore image structures, followed by an initial segmentation by applying MRF on the images and represented by a region adjacency graph (RAG). The proposed segmentation method has been evaluated using machine images. Relative to existing MRF-based methods, testing results have demonstrated that our proposed method substantially improves the segmentation performance.

Procedural Generation of Large-Scale Forests Using a Graph-Based Neutral Landscape Model

Specifying the positions and attributes of plants (e.g., species, size, and height) during the procedural generation of large-scale forests in virtual geographic environments is challenging, especially when reflecting the characteristics of vegetation distributions. To address this issue, a novel graph-based neutral landscape model (NLM) is proposed to generate forest landscapes with varying compositions and configurations. Our model integrates a set of class-level landscape metrics and generates more realistic and variable landscapes compared with existing NLMs controlled by limited global-level landscape metrics. To produce patches with particular sizes and shapes, a region adjacency graph is transformed from a cluster map that is generated based upon percolation theory; subsequently, optimal neighboring nodes in the graph are merged under restricted growth conditions from a source node. The locations of seeds are randomly placed and their species are classified according to the generated forest landscapes to obtain the final tree distributions. The results demonstrate that our method can generate realistic vegetation distributions representing different spatial patterns of species with a time efficiency that satisfies the requirements for constructing large-scale virtual forests.

Image Segmentation Using Linked Mean-Shift Vectors and Global/Local Attributes

This paper proposes novel noniterative mean-shift-based image segmentation that uses global and local attributes. The existing mean-shift-based methods use a fixed range bandwidth, and hence their accuracy is dependent on the range spectrum of an image. To resolve this dependency, this paper proposes to modify the range kernel in the mean-shift process to be anisotropic. The modification is conducted using a global attribute defined as the range covariance matrix of the image. Further, to alleviate oversegmentation, the proposed method merges the segments having similar local attributes more aggressively than other segments. The local attribute for each segment is defined as the sum of the variances of the chromatic components. Finally, to expedite the processing, the proposed method uses a region adjacency graph (RAG) for the merging process, thus differing from the existing linked mean-shift-based methods. In the experiments on the Berkeley segmentation data set, the use of the global and local attributes improved segmentation accuracy; the proposed method outperformed the state-of-the-art linked mean-shift-based method by showing an improvement of 2.15%, 3.16%, 3.32%, and 1.90% in probability rand index, segmentation covering, variation of information, and F-measure, respectively. Further, compared with the benchmark method, which uses the dilating and merging scheme, the proposed method improved the speed of the merging process 42 times by applying the RAG.

Multiple Moving Object Detection From UAV Videos Using Trajectories of Matched Regional Adjacency Graphs

Image registration has been long used as a basis for the detection of moving objects. Registration techniques attempt to discover correspondences between consecutive frame pairs based on image appearances under rigid and affine transformations. However, spatial information is often ignored, and different motions from multiple moving objects cannot be efficiently modeled. Moreover, image registration is not well suited to handle occlusion that can result in potential object misses. This paper proposes a novel approach to address these problems. First, segmented video frames from unmanned aerial vehicle captured video sequences are represented using region adjacency graphs of visual appearance and geometric properties. Correspondence matching (for visible and occluded regions) is then performed between graph sequences by using multigraph matching. After matching, region labeling is achieved by a proposed graph coloring algorithm which assigns a background or foreground label to the respective region. The intuition of the algorithm is that background scene and foreground moving objects exhibit different motion characteristics in a sequence, and hence, their spatial distances are expected to be varying with time. Experiments conducted on several DARPA VIVID video sequences as well as self-captured videos show that the proposed method is robust to unknown transformations, with significant improvements in overall precision and recall compared to existing works.

A novel Markov random field model based on region adjacency graph for T1 magnetic resonance imaging brain segmentation

ABSTRACTTissue segmentation in magnetic resonance brain scans is the most critical task in different aspects of brain analysis. Because manual segmentation of brain magnetic resonance imaging (MRI) images is a time‐consuming and labor‐intensive procedure, automatic image segmentation is widely used for this purpose. As Markov Random Field (MRF) model provides a powerful tool for segmentation of images with a high level of artifacts, it has been considered as a superior method. But because of the high computational cost of MRF, it is not appropriate for online processing. This article has proposed a novel method based on a proper combination of MRF model and watershed algorithm in order to alleviate the MRF's drawbacks. Results illustrate that the proposed method has a good ability in MRI image segmentation, and also decreases the computational time effectively, which is a valuable improvement in the online applications. © 2017 Wiley Periodicals, Inc. Int J Imaging Syst Technol, 27, 78–88, 2017

Contextual aerial image categorization using codebook

Effective categorization of the millions of aerial images from unmanned planes is a useful technique with several important applications. Previous methods on this task usually encountered such problems: (1) it is hard to represent the aerial images’ topologies efficiently, which are the key feature to distinguish the arial images rather than conventional appearance, and (2) the computational load is usually too high to build a realtime image categorization system. Addressing these problems, this paper proposes an efficient and effective aerial image categorization method based on a contextual topological codebook. The codebook of aerial images is learned with a multitask learning framework. The topology of each aerial image is represented with the region adjacency graph (RAG). Furthermore, a codebook containing topologies is learned by jointly modeling the contextual information, based on the extracted discriminative graphlets. These graphlets are integrated into a Bag-of-Words (BoW) representation for predicting aerial image categories. Contextual relation among local patches are taken into account in categorization to yield high categorization performance. Experimental results show that our approach is both effective and efficient.

Multilayer semantic segmentation of remote-sensing imagery using a hybrid object-based Markov random field model

ABSTRACTHigh spatial resolution (HR) remote-sensing image usually contains hierarchical semantic information. Many supervised methods have been developed to interpret this information through data training. In this article, without data training, a hybrid object-based Markov random field (HOMRF) model is proposed for multilayer semantic segmentation of remote-sensing images. In this method, label fields of different semantic layers are defined on the same region adjacency graph (RAG) of a given image, and a hybrid framework is suggested to capture and utilize the interactions within and between semantic layers by label fields. Namely a new transition probability matrix is introduced into the energy functions of label fields for describing the semantic context between layers, and the multilevel logistic model is employed to describe the interactions within the same layer. A principled probabilistic inference is developed to determine the optimal solution of the proposed method by iteratively updating each label field until convergence. The computational complexity of the proposed model is , where is the number of classes in all of the layers, is the number of sites in the probability graph of the MRF model, and is the number of iterations. Experimental results from various remote-sensing images demonstrate that the proposed method can produce higher segmentation accuracy than state-of-the-art MRF-based methods.

Segmentation of the Synthetic Aperture Radar Image Using the Watershed Transformation and Region Merging Technique

This paper presents an efficient algorithm that segments the synthetic aperture radar (SAR) image into several homogeneous regions by combining edgeand region-based information. A multi-direction ratio edge detector is used to capture the locally directional variation information of a SAR image and construct the ratio edge strength map (RESM) of the SAR image. Watershed transformation is performed on the thresholded RESM to obtain an over-segmentation result. An efficient hierarchical region merging procedure based on the region adjacency graph representation of the image regions is proposed. A novel dissimilarity measurement between the two adjacent regions is derived from the multiplicative noise speckle model of the multi-look SAR image. Experimental results and the performance evaluation show that the proposed method outperforms two widely used SAR image segmentation approaches.

Matching images based on consistency graph and region adjacency graphs

The image matching methods based on regions have many advantages over the point matching techniques, and the most charming one is that once region being matched, all pixels are matched in theory. It would benefit many applications, such as object retrieval, stereo corresponding, semantic understanding a scene, object tracking. This paper proposes a new region matching algorithm based on consistency graph and region adjacency graphs. Firstly, the segmented images are transformed into region adjacency graphs, and the potential region pairs and the potential edge segment pairs are packaged in a consistency graph. Since the rightly matched pair always is accompanied by harmonious neighbourhoods, the right correspondences tend to cluster together, and the error corresponding relationship should have few chances to connect to any compatible neighbourhood. Thus, the solution space is greatly reduced and the corresponding relationship can be found in a polynomial computational complexity just by a simple method, such as seed-growth method. To the best of our knowledge, the method is the first one to match two images by region adjacency graphs and find the corresponding relationship in a polynomial computational complexity. Experiments on the existing benchmark show that the proposed method could quickly find the right corresponding relationship between images with illumination, rotation and affine transformation.

Edge-Guided Image Object Detection in Multiscale Segmentation for High-Resolution Remotely Sensed Imagery

A new segmentation approach for high-resolution remotely sensed imagery that combines the global edge and region information is developed from a new scheme to monitor the best conditions for each growing object to obtain the corresponding meaningful image object during multiscale analysis. The approach, which is an extension of the image object detection approach, includes new algorithms for determination of region-growing criteria, edge-guided image object detection, and assessment of edges. The method consists of two stages: In the first stage, edges are acquired from edge detection with embedded confidence and stored in an R-tree, and initial objects are segmented by eCognition and organized in the region adjacency graph; in the second stage, meaningful image objects are obtained by incorporating multiscale segmentation and analyzing the edge completeness curve. The evaluation results of edge completeness are obtained within the process of multiscale segmentation, and the assessment for the segmentation results shows its merit in coastal remote sensing. Images containing plenty of weak edges or distributing scene objects with various sizes and shapes can fully embody the strength of this method.

Semantic Segmentation of Remote Sensing Imagery Using Object-Based Markov Random Field Model With Regional Penalties

This paper proposes a novel object-based Markov random field model (OMRF) for semantic segmentation of remote sensing images. First, the method employs the region size and edge information to build a weighted region adjacency graph (WRAG) for capturing the complicated interactions among objects. Thereafter, aimed at modeling object interactions in the OMRF, the size and edge information are further introduced into the Gibbs joint distribution of the random field as regional penalties. Finally, the semantic segmentation is achieved through a principled probabilistic inference of the OMRF with regional penalties. The proposed method is compared with other MRF-based methods and some state-of-the-art methods. Experiments are conducted on a series of synthetic and real-world images. Segmentation results demonstrate that our method provides better performance (an accuracy improvement about 3%). Moreover, we further discuss the application of the proposed method for classification.

Large-Scale Aerial Image Categorization Using a Multitask Topological Codebook.

Fast and accurately categorizing the millions of aerial images on Google Maps is a useful technique in pattern recognition. Existing methods cannot handle this task successfully due to two reasons: 1) the aerial images' topologies are the key feature to distinguish their categories, but they cannot be effectively encoded by a conventional visual codebook and 2) it is challenging to build a realtime image categorization system, as some geo-aware Apps update over 20 aerial images per second. To solve these problems, we propose an efficient aerial image categorization algorithm. It focuses on learning a discriminative topological codebook of aerial images under a multitask learning framework. The pipeline can be summarized as follows. We first construct a region adjacency graph (RAG) that describes the topology of each aerial image. Naturally, aerial image categorization can be formulated as RAG-to-RAG matching. According to graph theory, RAG-to-RAG matching is conducted by enumeratively comparing all their respective graphlets (i.e., small subgraphs). To alleviate the high time consumption, we propose to learn a codebook containing topologies jointly discriminative to multiple categories. The learned topological codebook guides the extraction of the discriminative graphlets. Finally, these graphlets are integrated into an AdaBoost model for predicting aerial image categories. Experimental results show that our approach is competitive to several existing recognition models. Furthermore, over 24 aerial images are processed per second, demonstrating that our approach is ready for real-world applications.

Automatic image segmentation using grid-based Ncut and RAG

Irrefutably normalised cuts (Ncut) is one of the most popular segmentation algorithms in computer vision. Inspite of its computational complexity, it has been applied to a wide range of segmentation tasks in computer vision for achieving good segmentation results. The difficulty in Ncut algorithm lies in handling images with high resolution and a predefined number of segments which results in increasing the complexity of the algorithm. In this paper, the drawbacks of the Ncut algorithm are studied and an efficient methodology for automatic image segmentation is proposed using grid based Ncut and merging of regions using region adjacency graph (RAG). The proposed algorithm is successful in two aspects, one in reducing the time complexity and second in automatically determining the number of image segments. The proposed methodology is implemented in three phases which includes preprocessing, segmentation and merging phases. The datasets used for experimentation are NSU, Berkeley, Corel 5K datasets and a few images from our own collection. Various experimental results performed on several images proved to be efficient in reducing the computational complexity and thus increasing the performance of segmentation when compared to standard Ncut algorithms.

Fast SAR Image Segmentation via Merging Cost With Relative Common Boundary Length Penalty

In this paper, a region-merging-based method is proposed for fast segmentation of amplitude-format synthetic aperture radar (SAR) images. It combines the existing fast initial partition by applying the watershed transform to the thresholded ratio edge strength map with fast region merging by using a new merging cost with relative common boundary length penalty (RCBLP) and the nearest neighbor graph (NNG) for fast search minimal edge on a region adjacency graph (RAG). A new statistical similarity measure, which is a scale-invariant and approximately constant false alarm rate with respect to region sizes, is proposed and combined with an RCBLP term to form a new merging cost. The region-merging process starting from the initial partition is fast implemented by means of the RAG and NNG. Several quantitative indexes in optical image segmentation assessment are borrowed for quantitative assessment of segmentation quality. Experiments to synthetic and real SAR images are reported. The results show that the proposed method is fast and attains higher quality segmentation results than the two recent state-of-the-art methods.

Unsupervised synthetic aperture radar image segmentation with superpixels in independent space based on independent component analysis

Synthetic aperture radar (SAR) image segmentation is a challenging problem in recent years because of the speckle noise. An unsupervised SAR image segmentation with superpixels by independent component analysis (ICA) is proposed. ICA independent space is proposed to represent SAR images for feature extraction effectively. First, the SAR image is divided into small regions by mean-shift algorithm and then those regions are merged in region adjacent graph and full-connected graph based on the Mining Spanning Tree theory, which balances the speed and quality of segmentation. Finally, experiments on X-band TerraSAR images and comparisons with simple linear iterative clustering and graph-cut illustrate the excellent performance of the new method.

Eikonal based region growing for superpixels generation: Application to semi-supervised real time organ segmentation in CT images

We describe a semi-supervised organ segmentation method for Computed Tomography images. In a first step, a dense oversegmentation of the image is created with an Eikonal-based algorithm. The proposed superpixel algorithm ourperforms state-of-the-art algorithms on classical metrics. In a second step, the semi-supervised segmentation is performed on the underlying Region Adjacency Graph created from the oversegmentation. As the complexity is greatly reduced, the organ segmentation can be performed in real-time.

Cell detection and segmentation using correlation clustering.

Cell detection and segmentation in microscopy images is important for quantitative high-throughput experiments. We present a learning-based method that is applicable to different modalities and cell types, in particular to cells that appear almost transparent in the images. We first train a classifier to detect (partial) cell boundaries. The resulting predictions are used to obtain superpixels and a weighted region adjacency graph. Here, edge weights can be either positive (attractive) or negative (repulsive). The graph partitioning problem is then solved using correlation clustering segmentation. One variant we newly propose here uses a length constraint that achieves state-of-art performance and improvements in some datasets. This constraint is approximated using non-planar correlation clustering. We demonstrate very good performance in various bright field and phase contrast microscopy experiments.

Structured representations in a content based image retrieval context

Here, we propose an automatic system to annotate and retrieve images. We assume that regions in an image can be described using a vocabulary of blobs. Blobs are generated from image features using clustering. Features are locally extracted on regions to capture Color, Texture and Shape information. Regions are processed by an efficient segmentation algorithm. Images are structured into a region adjacency graph to consider spatial relationships between regions. This representation is used to perform a similarity search into an image set. Hence, the user can express his need by giving a query image, and thereafter receiving as a result all similar images. Our graph based approach is benchmarked to conventional Bag of Words methods. Results tend to reveal a good behavior in classification of our graph based solution on two publicly available databases. Experiments illustrate that a structural approach requires a smaller vocabulary size to reach its best performance.

Discovering Discriminative Graphlets for Aerial Image Categories Recognition

Recognizing aerial image categories is useful for scene annotation and surveillance. Local features have been demonstrated to be robust to image transformations, including occlusions and clutters. However, the geometric property of an aerial image (i.e., the topology and relative displacement of local features), which is key to discriminating aerial image categories, cannot be effectively represented by state-of-the-art generic visual descriptors. To solve this problem, we propose a recognition model that mines graphlets from aerial images, where graphlets are small connected subgraphs reflecting both the geometric property and color/texture distribution of an aerial image. More specifically, each aerial image is decomposed into a set of basic components (e.g., road and playground) and a region adjacency graph (RAG) is accordingly constructed to model their spatial interactions. Aerial image categories recognition can subsequently be casted as RAG-to-RAG matching. Based on graph theory, RAG-to-RAG matching is conducted by comparing all their respective graphlets. Because the number of graphlets is huge, we derive a manifold embedding algorithm to measure different-sized graphlets, after which we select graphlets that have highly discriminative and low redundancy topologies. Through quantizing the selected graphlets from each aerial image into a feature vector, we use support vector machine to discriminate aerial image categories. Experimental results indicate that our method outperforms several state-of-the-art object/scene recognition models, and the visualized graphlets indicate that the discriminative patterns are discovered by our proposed approach.