Number Of Superpixels Research Articles

Superpixel based recursive least-squares method for lossless compression of hyperspectral images

Filtering based compression methods have become a popular research topic in lossless compression of hyperspectral images. Recursive least squares (RLS) based prediction methods provide better decorrelation performance among the filtering based methods. In this paper, two superpixel segmentation based RLS methods, namely SuperRLS and B-SuperRLS, are investigated for lossless compression of hyperspectral images. The proposed methods present a novel parallelization approach for RLS based prediction method. In the first step of SuperRLS, superpixel segmentation is applied to hyperspectral image. Afterwards, hyperspectral image is partitioned into multiple small regions according to the superpixel boundaries. Each region is predicted with RLS method in parallel, and prediction residuals are encoded via arithmetic encoder. Additionally, superpixel based prediction approach provides region of interest compression capability. B-SuperRLS, which is bimodal version of SuperRLS, evaluates both spectral and spatio-spectral correlations for prediction. The performance of the proposed methods are exhaustively analysed in terms of superpixel number, input vector length and number of parallel nodes, used in the prediction. Experimental results show that the proposed parallel architecture dramatically reduces the computation time, and achieves lower bit-rate performances among the state-of-the-art methods.

Extinction Profiles Fusion for Hyperspectral Images Classification

An extinction profile (EP) is an effective spatial-spectral feature extraction method for hyperspectral images (HSIs), which has recently drawn much attention. However, the existing methods utilize the EPs in a stacking way, which is hard to fully explore the information in EPs for HSI classification. In this paper, a novel fusion framework termed EPs-fusion (EPs-F) is proposed to exploit the information within and among EPs for HSI classification. In general, EPs-F includes the following two stages. In the first stage, by extracting the EPs from three independent components of an HSI, three complementary groups of EPs can be constructed. For each EP, an adaptive superpixel-based composite kernel strategy is proposed to explore the spatial information within an EP. The weights to create the composite kernel and the number of superpixels are automatically determined based on the spatial information of each EP. In the second stage, since the different EPs contain highly complementary information, a simple yet effective decision fusion method is further applied to obtain the final classification result. Experiments on three real HSI data sets verify the qualitative and quantitative superiority of the proposed EPs-F method over several state-of-the-art HSI classifiers.

An iterative approach for obtaining multi-scale superpixels based on stochastic graph contraction operations

Superpixels have many applications in visual information processing, and can be used to reduce redundant information of an image, as well as the computational complexity of other expensive tasks (e.g., image segmentation). In this work, an iterative hierarchical stochastic graph contraction (IHSGC) method for multi-scale superpixels generation is proposed. A stochastic strategy is used to generate multi-scale superpixels, and each superpixel is represented by a hierarchical tree and describes an image patch at fine and coarse scales simultaneously. The proposed method consists of two main steps. The first step initializes the method based on a multi-channel unsupervised stochastic over-segmentation at the pixel level. The proposed over-segmentation scheme actually performs hierarchical stochastic clustering of visual features (i.e. pixels, image patches, and potentially can be applied to other visual features as well), while preserving the local spatial relationships across different scales. The second step consists of an iterative hierarchical stochastic graph contraction method. Coarser scales are generated by graph contractions until the desired number of superpixels is obtained. The experimental results based on the popular Berkeley segmentation databases BSDS300 and BSDS500 suggest that the proposed approach potentially can perform better than comparative state-of-the-art methods in terms of boundary recall and under-segmentation error.

Fast PET Scan Tumor Segmentation Using Superpixels, Principal Component Analysis and K-Means Clustering

Positron Emission Tomography scan images are extensively used in radiotherapy planning, clinical diagnosis, assessment of growth and treatment of a tumor. These all rely on fidelity and speed of detection and delineation algorithm. Despite intensive research, segmentation has remained a challenging problem due to the diverse image content, resolution, shape, and noise. This paper presents a fast positron emission tomography tumor segmentation method using superpixels. Principal component analysis is applied on the superpixels and their average value. The distance vector of each superpixel from the average is computed in the principal components coordinate system. Finally, k-means clustering is applied on the distance vector to recognize tumor and non-tumor superpixels. The proposed approach is implemented in MATLAB 2016A, and promising accuracy with execution time of 2.35 ± 0.26 s is achieved. Fast execution time is achieved since the number of superpixels, and the size of distance vector on which clustering was done are low compared to the number of pixels in the image.

Salient object detection based on eye tracking data

Salient object detection aims to detect the region of interest in an image and provides an efficient solution for image semantic understanding. Eye tracking data can accurately track where human interested in an image, which has been introduced for image processing recently. In this paper, we propose a new saliency detection model with a combination of superpixel segmentation and eye tracking data. Eye tracking data are firstly introduced for reducing the number of superpixels and speeding up the calculation. And then, we give a strategy for training sets selection and construct a learning model to detect salient object. Finally, the feature contrast and multi-scale strategy are combined to propose an integrated model and get the refinement saliency map. We compare our method to the 10 state-of-the-art methods on the two public image datasets. These results show that our method outperforms existing methods and eye tracking technology is a highly promising tools for saliency object detection.

Two-level superpixel and feedback based visual object tracking

While numerous superpixel-based tracking algorithms have been proposed and demonstrated successfully, there still remain some challenges, such as determining the number of superpixels, mining and exploiting the structural information of superpixels and handling the drifts. In this paper, we propose a tracking method with two-level superpixels and a novel update strategy based on feedback to deal with the challenges mentioned above. Firstly, Bilateral filter is introduced to filter out outliers and improve the boundary capability of object as well as segmentation of superpixels. Then two-level superpixel is proposed to determine superpixel number automatically through iterating instead of setting superpixel number empirically which affects the robustness of tracking algorithm. Moreover, a novel measuring method which considers color similarity and relative positions of superpixels is proposed to make a better use of structural information of superpixels and improve tracking performance by adding relative position of superpixels into the appearance model. Finally, a feedback based update strategy is presented to handle drifts existing in tracking by calculating the adaptation of appearance model and updating the parameters like superpixel number and relative position of superpixels. Experiments on challenging sequences and comparisons to state-of-the-art methods demonstrate the feasibility and effectiveness of the proposed tracking algorithm.

Multiscale Superpixel-Based Sparse Representation for Hyperspectral Image Classification

Recently, superpixel segmentation has been proven to be a powerful tool for hyperspectral image (HSI) classification. Nonetheless, the selection of the optimal superpixel size is a nontrivial task. In addition, compared with single-scale superpixel segmentation, the same image segmented on a different scale can obtain different structure information. To overcome such a drawback also utilizing the structural information, a multiscale superpixel-based sparse representation (MSSR) algorithm for the HSI classification is proposed. Specifically, a modified segmentation strategy of multiscale superpixels is firstly applied on the HSI. Once the superpixels on different scales are obtained, the joint sparse representation classification is used to classify the multiscale superpixels. Furthermore, majority voting is utilized to fuse the labels of different scale superpixels and to obtain the final classification result. Two merits are realized by the MSSR. First, multiscale information fusion can more effectively explore the spatial information of HSI. Second, in the multiscale superpixel segmentation, except for the first scale, the superpixel number on a different scale for different HSI datasets can be adaptively changed based on the spatial complexity of the corresponding HSI. Experiments on four real HSI datasets demonstrate the qualitative and quantitative superiority of the proposed MSSR algorithm over several well-known classifiers.

A Practical GrabCut Color Image Segmentation Based on Bayes Classification and Simple Linear Iterative Clustering

In order to solve the segmentation degradation phenomenon when the number of super pixels is low, we propose a novel color image segmentation algorithm based on GrabCut. The method integrates Bayes classification with simple linear iterative clustering (SLIC) and then use the GrabCut method to obtain the segmentation. The SLIC is applied to cluster the features of a color image and integrated it into the GrabCut framework to overcome the problem of the image segmentation deterioration when the number of super pixels is low. In addition, we extend the Gaussian mixture model (GMM) to SLIC features and GMM based on SLIC is constructed to describe the energy function. The color clustering can be suitably integrated into the GrabCut framework and fused with the color feature to achieve more superior image segmentation performance than the original GrabCut method. For easier implementation and more efficient computation, the Bayes classification is chosen for reconstruction of the simplified graph cut model instead of the original graph cut based on the SLIC model. The min-cut algorithm technique served as the division measure in the simplified image space for more discriminating power. A classification strategy is presented, to effectively adjust the energy function so that the Bayes classification and SLIC features are efficiently integrated to achieve more robust segmentation performance. Finally, boundary optimization is proposed to dramatically reduce the boundary roughness of the GrabCut algorithm with satisfactory segmentation accuracy. As a practical application, the superior performance of our proposed method was demonstrated through a large number of comparative tests.

Multiscale and Multifeature Normalized Cut Segmentation for High Spatial Resolution Remote Sensing Imagery

In this paper, a framework for multiscale and multifeature normalized cut (MMNCut) segmentation is proposed for high spatial resolution (HSR) remote sensing images. Normalized cuts (NCuts), as a widely used segmentation method for natural images, can obtain a globally optimized segmentation result corresponding to the optimized partitions of a graph. However, it is difficult to apply the traditional NCuts directly to HSR images because of the huge computational complexity and the diversity of the characteristics of the land covers. In order to solve these problems, the proposed MMNCuts builds a multiscale graph based on superpixels, which can provide powerful grouping cues to guide the segmentation. Generated by different algorithms with varying parameters, superpixels can capture diverse and multiscale visual patterns of HSR images. In addition, the newly constructed graph integrates the multiscale information by considering various connection relationships. Meanwhile, the successful integration of the multifeature cues, including the spectral information, texture information, and structure information, from a large number of superpixels, helps to enhance the expression ability of the graph. Computationally, this leads to a much more efficient algorithm than the traditional NCuts, and in effect, the proposed method achieves a significantly better performance than the traditional approaches. The experimental results with three HSR image data sets demonstrate that the proposed MMNCut algorithm shows a competitive performance in both qualitative and quantitative evaluations when compared with the other state-of-the-art segmentation algorithms for HSR images.

Automatic superpixel generation algorithm based on a quadric error metric in 3D space

As a preprocessing step for computer vision tasks, research on generating superpixels of an image has been widely conducted recently. In this paper, we propose a fully automatic superpixel generation algorithm by simplifying the 3D triangle mesh modeled from a 2D input image. The simplification is performed based on a modified quadric error metric (QEM) method. The pipeline of our algorithm is simple. Given an image, we first turn it into a 2D triangle mesh and then lift the mesh to 3D based on the gray values of the image. We then simplify the 3D mesh based on the modified QEM method which encodes the features in the image intrinsically to ensure feature-aware superpixels. After obtaining the simplified mesh, we map it back to the 2D image to generate triangular-shaped superpixels. Our algorithm is fully automatic. The number of superpixels can be controlled intuitively. Experimental results demonstrate the effectiveness of our approach.

A Simple Algorithm of Superpixel Segmentation with Boundary Constraint

As one of the most popular image oversegmentations, superpixel has been commonly used as supporting regions for primitives to reduce computations in various computer vision tasks. In this paper, we propose a novel superpixel segmentation approach based on a distance function that is designed to balance among boundary adherence, intensity homogeneity, and compactness (COM) characteristics of the resulting superpixels. Given an expected number of superpixels, our method begins with initializing the superpixel seed positions to obtain the initial labels of pixels. Then, we optimize the superpixels iteratively based on the defined distance measurement. We update the positions and intensities of superpixel seeds based on the three-sigma rule. The experimental results demonstrate that our algorithm is more effective and accurate than previous superpixel methods and achieves a comparable tradeoff between superpixel COM and adherence to object boundaries.

Image Categorization by Learning a Propagated Graphlet Path.

Spatial pyramid matching is a standard architecture for categorical image retrieval. However, its performance is largely limited by the prespecified rectangular spatial regions when pooling local descriptors. In this paper, we propose to learn object-shaped and directional receptive fields for image categorization. In particular, different objects in an image are seamlessly constructed by superpixels, while the direction captures human gaze shifting path. By generating a number of superpixels in each image, we construct graphlets to describe different objects. They function as the object-shaped receptive fields for image comparison. Due to the huge number of graphlets in an image, a saliency-guided graphlet selection algorithm is proposed. A manifold embedding algorithm encodes graphlets with the semantics of training image tags. Then, we derive a manifold propagation to calculate the postembedding graphlets by leveraging visual saliency maps. The sequentially propagated graphlets constitute a path that mimics human gaze shifting. Finally, we use the learned graphlet path as receptive fields for local image descriptor pooling. The local descriptors from similar receptive fields of pairwise images more significantly contribute to the final image kernel. Thorough experiments demonstrate the advantage of our approach.

Normalized cut-based saliency detection by adaptive multi-level region merging.

Existing salient object detection models favor over-segmented regions upon which saliency is computed. Such local regions are less effective on representing object holistically and degrade emphasis of entire salient objects. As a result, the existing methods often fail to highlight an entire object in complex background. Toward better grouping of objects and background, in this paper, we consider graph cut, more specifically, the normalized graph cut (Ncut) for saliency detection. Since the Ncut partitions a graph in a normalized energy minimization fashion, resulting eigenvectors of the Ncut contain good cluster information that may group visual contents. Motivated by this, we directly induce saliency maps via eigenvectors of the Ncut, contributing to accurate saliency estimation of visual clusters. We implement the Ncut on a graph derived from a moderate number of superpixels. This graph captures both intrinsic color and edge information of image data. Starting from the superpixels, an adaptive multi-level region merging scheme is employed to seek such cluster information from Ncut eigenvectors. With developed saliency measures for each merged region, encouraging performance is obtained after across-level integration. Experiments by comparing with 13 existing methods on four benchmark datasets, including MSRA-1000, SOD, SED, and CSSD show the proposed method, Ncut saliency, results in uniform object enhancement and achieves comparable/better performance to the state-of-the-art methods.

Pixel clustering for color image segmentation

Image segmentation using a hierarchical sequence of piecewise constant approximations that minimally differ from the original image in terms of the total squared error is discussed. It is proposed to obtain these approximations by two combined clustering and segmentation methods based on clustering image pixels using Ward's method. In the first method, the number of segments in clusters is reduced in the course of hierarchical clustering by reclassifying pixels from one cluster to another. In the second method, a limited number of superpixels representing connected segments of the image are formed by enlarging source pixels, and then the superpixels are clusterized by Ward's method. To decompose the image into superpixels, the segmentation quality is improved while preserving the number of segments. As a result, a noticeable improvement in the quality of image approximations is achieved, and their invariant encoding gives a marking of the image for subsequent object detection.

Correlation between depth and area of retinal nerve fiber layer defect as measured by spectral domain optical coherence tomography.

To evaluate the correlation between the depth and area of retinal nerve fiber layer (RNFL) defect, as measured on an RNFL map of spectral-domain optical coherence tomography (SD-OCT). The RNFL of 472 glaucoma subjects and of 217 healthy subjects was imaged by an SD-OCT. RNFL defect depth and area on the RNFL map were expressed as an RNFL defect depth percentage index (RDPI) and an RNFL defect area index (RDAI), respectively, according to the following two formulas: 100×[1-{summation of thicknesses of RNFL defects/summation of thicknesses of upper 95th percentile range of age-matched healthy subjects in areas corresponding to defects}]; 100×[number of superpixels of RNFL defects/(46 × 46-superpixels inside optic disc or β zone parapapillary atrophy)]. The best-fitting model describing the relationship between the two parameters was derived by fractional polynomial analysis. Logarithmic fit was determined to be the best-fitting model in describing the relationship of the RDPI against the RDAI (y = 53.4 + 3.7 ln(x) and y = 50.9 + 5.9 ln(x) in superior and inferior hemifields, respectively). The expected RDAIs at the point where the RDPI and RDAI rates of change were the same were 3.7 and 5.9 %; the corresponding upper 95 % confidence interval limits of the RDPI 59.0 and 61.8 % in the superior and inferior hemifields, respectively. The correlation between the RNFL defect depth and area, as derived from the RNFL map, was best described by the logarithmic fit. Changes were more marked in depth than in area, especially for mild localized defects.

Recognising lepidopteran images based on locality-constrained linear coding and SVM

In this paper, a novel lepidopteran insect images recognition method is proposed. The captured insect images are first preprocessed to separate the foreground from the complicated background. Then, two wings are cut out from the insect, calibrated and segmented into a number of superpixels. The values of l, a, b, x, y of each superpixel are calculated as feature descriptors. Those descriptors are encoded with locality-constrained linear coding and then pooled into a feature vector. Finally, the multi-class linear support vector machine is used for classification. The approach is testified in a data-set including 576 images of insect specimens from 10 different lepidopteran species and the recognition accuracy is over 99% in average. The experimental results suggested that the proposed method performs well in lepidopteran insect species recognition.

Performance evaluation of simple linear iterative clustering algorithm on medical image processing.

Simple Linear Iterative Clustering (SLIC) algorithm is increasingly applied to different kinds of image processing because of its excellent perceptually meaningful characteristics. In order to better meet the needs of medical image processing and provide technical reference for SLIC on the application of medical image segmentation, two indicators of boundary accuracy and superpixel uniformity are introduced with other indicators to systematically analyze the performance of SLIC algorithm, compared with Normalized cuts and Turbopixels algorithm. The extensive experimental results show that SLIC is faster and less sensitive to the image type and the setting superpixel number than other similar algorithms such as Turbopixels and Normalized cuts algorithms. And it also has a great benefit to the boundary recall, the robustness of fuzzy boundary, the setting superpixel size and the segmentation performance on medical image segmentation.

Three-Dimensional Spectral-Domain Optical Coherence Tomography Data Analysis for Glaucoma Detection

PurposeTo develop a new three-dimensional (3D) spectral-domain optical coherence tomography (SD-OCT) data analysis method using a machine learning technique based on variable-size super pixel segmentation that efficiently utilizes full 3D dataset to improve the discrimination between early glaucomatous and healthy eyes.Methods192 eyes of 96 subjects (44 healthy, 59 glaucoma suspect and 89 glaucomatous eyes) were scanned with SD-OCT. Each SD-OCT cube dataset was first converted into 2D feature map based on retinal nerve fiber layer (RNFL) segmentation and then divided into various number of super pixels. Unlike the conventional super pixel having a fixed number of points, this newly developed variable-size super pixel is defined as a cluster of homogeneous adjacent pixels with variable size, shape and number. Features of super pixel map were extracted and used as inputs to machine classifier (LogitBoost adaptive boosting) to automatically identify diseased eyes. For discriminating performance assessment, area under the curve (AUC) of the receiver operating characteristics of the machine classifier outputs were compared with the conventional circumpapillary RNFL (cpRNFL) thickness measurements.ResultsThe super pixel analysis showed statistically significantly higher AUC than the cpRNFL (0.855 vs. 0.707, respectively, p = 0.031, Jackknife test) when glaucoma suspects were discriminated from healthy, while no significant difference was found when confirmed glaucoma eyes were discriminated from healthy eyes.ConclusionsA novel 3D OCT analysis technique performed at least as well as the cpRNFL in glaucoma discrimination and even better at glaucoma suspect discrimination. This new method has the potential to improve early detection of glaucomatous damage.

슈퍼픽셀과 FCM을 이용한 클러스터 초기값 설정 및 칼라영상분할

본 논문에서는 슈퍼픽셀과 FCM을 이용한 클러스터 초기값 설정방법과 이를 사용한 칼라영상분할을 연구한다. 클러스터링을 이용한 대표적인 칼라영상분할 방법으로 Fuzzy C-menas (FCM) 알고리즘을 많이 사용한다. FCM은 하나의 데이터가 각 클러스터에 서로 다른 소속도를 갖도록 한다. 그러나 FCM은 초기값 설정에 따라 국부적인 수렴문제가 발생한다. 따라서 초기값 설정문제는 매우 중요한데 본 연구에서는 슈퍼픽셀을 이용하여 클러스터의 초기값을 구하는 방법을 제안한다. 슈퍼픽셀은 원 영상에서 특성이 비슷한 화소들의 묶음으로 표현되는데 먼저 원 영상으로부터 슈퍼픽셀을 구하고 이를 <TEX>$La^*b^*$</TEX> 칼라특징공간에 투영하여 클러스터 초기값을 구한다. 제안방법에서 슈퍼픽셀의 수는 원영상의 화소 수보다 일반적으로 매우 적어서 클러스터 초기값 설정을 위한 고속처리가 가능하다. 제안된 알고리즘의 성능평가를 위해 다양한 칼라영상을 사용하여 컴퓨터 모의실험을 수행하였으며 실험결과 제안방법이 기존방법에 비해 영상분할 성능이 우수함을 알 수 있었다. In this paper, a setting method of initial cluster centers and color image segmentation using superpixels and Fuzzy C-means(FCM) algorithm is proposed. Generally, the FCM can be widely used to segment color images, and an element is assigned to any cluster with each membership values in the FCM. However the algorithm has a problem of local convergence by determining the initial cluster centers. So the selection of initial cluster centers is very important, we proposed an effective method to determine the initial cluster centers using superpixels. The superpixels can be obtained by grouping of some pixels having similar characteristics from original image, and it is projected <TEX>$La^*b^*$</TEX> feature space to obtain the initial cluster centers. The proposed method can be speeded up because number of superpixels are extremely smaller than pixels of original image. To evaluate the proposed method, several color images are used for computer simulation, and we know that the proposed method is superior to the conventional algorithm by the experimental results.

Clinical Evaluation of the Proper Orthogonal Decomposition Framework for Detecting Glaucomatous Changes in Human Subjects

To evaluate the new proper orthogonal decomposition (POD) framework for detecting glaucomatous progression from HRT topographies of human subjects and compare it with HRT topographic change analysis (TCA). Of 267 eyes of 187 participants with > or =4 retinal tomographic examinations in the University of California, San Diego Diagnostic Innovations in Glaucoma Study (DIGS), 21 eyes were of longitudinally normal subjects and 36 eyes progressed by stereophotographs or visual field-guided progression analysis (progressors). All others were considered nonprogressing (nonprogressors; n = 210 eyes). POD parameters of Euclidean distance (L(2) norm), image Euclidean distance, and correlation were computed, and their area under receiver operating characteristic curves (AUC) in differentiating progressors from nonprogressors and normal subjects were compared to the TCA parameters of the number of superpixels with significant decrease in retinal height (red pixels), size of the largest cluster of red pixels (CSIZE), and CSIZE% of disc size, all within the optic disc margin. AUCs of the best performing POD L(2) norm and TCA red pixel parameters in differentiating progressors from normal subjects were both 0.86 and in differentiating progressors from nonprogressors were 0.68 and 0.64, respectively; the AUC differences were not statistically significant. The POD framework, which can detect and confirm glaucomatous changes in a single follow-up visit, provides a performance similar to that of TCA in differentiating progressors from normal subjects and nonprogressors.