Showing papers on "Thresholding published in 2016"
01 Jun 2016
TL;DR: The proposed OpenMax model significantly outperforms open set recognition accuracy of basic deep networks as well as deep networks with thresholding of SoftMax probabilities, and it is proved that the OpenMax concept provides bounded open space risk, thereby formally providing anopen set recognition solution.
Abstract: Deep networks have produced significant gains for various visual recognition problems, leading to high impact academic and commercial applications. Recent work in deep networks highlighted that it is easy to generate images that humans would never classify as a particular object class, yet networks classify such images high confidence as that given class – deep network are easily fooled with images humans do not consider meaningful. The closed set nature of deep networks forces them to choose from one of the known classes leading to such artifacts. Recognition in the real world is open set, i.e. the recognition system should reject unknown/unseen classes at test time. We present a methodology to adapt deep networks for open set recognition, by introducing a new model layer, OpenMax, which estimates the probability of an input being from an unknown class. A key element of estimating the unknown probability is adapting Meta-Recognition concepts to the activation patterns in the penultimate layer of the network. Open-Max allows rejection of "fooling" and unrelated open set images presented to the system, OpenMax greatly reduces the number of obvious errors made by a deep network. We prove that the OpenMax concept provides bounded open space risk, thereby formally providing an open set recognition solution. We evaluate the resulting open set deep networks using pre-trained networks from the Caffe Model-zoo on ImageNet 2012 validation data, and thousands of fooling and open set images. The proposed OpenMax model significantly outperforms open set recognition accuracy of basic deep networks as well as deep networks with thresholding of SoftMax probabilities.
1,034 citations
27 Jun 2016
TL;DR: In this article, a CNN-RNN framework is proposed to learn a joint image-label embedding to characterize the semantic label dependency as well as the image label relevance, and it can be trained end-to-end from scratch to integrate both information in a unified framework.
Abstract: While deep convolutional neural networks (CNNs) have shown a great success in single-label image classification, it is important to note that real world images generally contain multiple labels, which could correspond to different objects, scenes, actions and attributes in an image. Traditional approaches to multi-label image classification learn independent classifiers for each category and employ ranking or thresholding on the classification results. These techniques, although working well, fail to explicitly exploit the label dependencies in an image. In this paper, we utilize recurrent neural networks (RNNs) to address this problem. Combined with CNNs, the proposed CNN-RNN framework learns a joint image-label embedding to characterize the semantic label dependency as well as the image-label relevance, and it can be trained end-to-end from scratch to integrate both information in a unified framework. Experimental results on public benchmark datasets demonstrate that the proposed architecture achieves better performance than the state-of-the-art multi-label classification models.
933 citations
TL;DR: The proposed algorithm of applying the LAD filter on orientation scores (LAD-OS) outperforms most of the state-of-the-art methods and is capable of dealing with typically difficult cases like crossings, central arterial reflex, closely parallel and tiny vessels.
Abstract: This paper presents a robust and fully automatic filter-based approach for retinal vessel segmentation. We propose new filters based on 3D rotating frames in so-called orientation scores, which are functions on the Lie-group domain of positions and orientations $\mathbb {R}^{2} \rtimes S^{1}$ . By means of a wavelet-type transform, a 2D image is lifted to a 3D orientation score, where elongated structures are disentangled into their corresponding orientation planes. In the lifted domain $\mathbb {R}^{2} \rtimes S^{1}$ , vessels are enhanced by means of multi-scale second-order Gaussian derivatives perpendicular to the line structures. More precisely, we use a left-invariant rotating derivative (LID) frame, and a locally adaptive derivative (LAD) frame. The LAD is adaptive to the local line structures and is found by eigensystem analysis of the left-invariant Hessian matrix (computed with the LID). After multi-scale filtering via the LID or LAD in the orientation score domain, the results are projected back to the 2D image plane giving us the enhanced vessels. Then a binary segmentation is obtained through thresholding. The proposed methods are validated on six retinal image datasets with different image types, on which competitive segmentation performances are achieved. In particular, the proposed algorithm of applying the LAD filter on orientation scores (LAD-OS) outperforms most of the state-of-the-art methods. The LAD-OS is capable of dealing with typically difficult cases like crossings, central arterial reflex, closely parallel and tiny vessels. The high computational speed of the proposed methods allows processing of large datasets in a screening setting.
318 citations
Posted Content•
TL;DR: The proposed CNN-RNN framework learns a joint image-label embedding to characterize the semantic label dependency as well as the image- label relevance, and it can be trained end-to-end from scratch to integrate both information in a unified framework.
Abstract: While deep convolutional neural networks (CNNs) have shown a great success in single-label image classification, it is important to note that real world images generally contain multiple labels, which could correspond to different objects, scenes, actions and attributes in an image. Traditional approaches to multi-label image classification learn independent classifiers for each category and employ ranking or thresholding on the classification results. These techniques, although working well, fail to explicitly exploit the label dependencies in an image. In this paper, we utilize recurrent neural networks (RNNs) to address this problem. Combined with CNNs, the proposed CNN-RNN framework learns a joint image-label embedding to characterize the semantic label dependency as well as the image-label relevance, and it can be trained end-to-end from scratch to integrate both information in a unified framework. Experimental results on public benchmark datasets demonstrate that the proposed architecture achieves better performance than the state-of-the-art multi-label classification model
305 citations
TL;DR: The usefulness of surroundedness for eye fixation prediction by proposing a Boolean Map based Saliency model (BMS), based on a Gestalt principle of figure-ground segregation, which computes a saliency map by discovering surrounded regions via topological analysis of Boolean maps.
Abstract: We demonstrate the usefulness of surroundedness for eye fixation prediction by proposing a Boolean Map based Saliency model (BMS). In our formulation, an image is characterized by a set of binary images, which are generated by randomly thresholding the image's feature maps in a whitened feature space. Based on a Gestalt principle of figure-ground segregation, BMS computes a saliency map by discovering surrounded regions via topological analysis of Boolean maps. Furthermore, we draw a connection between BMS and the Minimum Barrier Distance to provide insight into why and how BMS can properly captures the surroundedness cue via Boolean maps. The strength of BMS is verified by its simplicity, efficiency and superior performance compared with 10 state-of-the-art methods on seven eye tracking benchmark datasets.
237 citations
TL;DR: The experimental results demonstrate that the proposed method can effectively reduce noise and be competitive with the current state-of-the-art denoising algorithms in terms of both quantitative metrics and subjective visual quality.
Abstract: Nonlocal self-similarity of images has attracted considerable interest in the field of image processing and has led to several state-of-the-art image denoising algorithms, such as block matching and 3-D, principal component analysis with local pixel grouping, patch-based locally optimal wiener, and spatially adaptive iterative singular-value thresholding. In this paper, we propose a computationally simple denoising algorithm using the nonlocal self-similarity and the low-rank approximation (LRA). The proposed method consists of three basic steps. First, our method classifies similar image patches by the block-matching technique to form the similar patch groups, which results in the similar patch groups to be low rank. Next, each group of similar patches is factorized by singular value decomposition (SVD) and estimated by taking only a few largest singular values and corresponding singular vectors. Finally, an initial denoised image is generated by aggregating all processed patches. For low-rank matrices, SVD can provide the optimal energy compaction in the least square sense. The proposed method exploits the optimal energy compaction property of SVD to lead an LRA of similar patch groups. Unlike other SVD-based methods, the LRA in SVD domain avoids learning the local basis for representing image patches, which usually is computationally expensive. The experimental results demonstrate that the proposed method can effectively reduce noise and be competitive with the current state-of-the-art denoising algorithms in terms of both quantitative metrics and subjective visual quality.
228 citations
TL;DR: The proposed method is based on the synchrosqueezed continuous wavelet transform (SS-CWT) and custom thresholding of single-channel data and incorporates a detection procedure that uses the thresholded wavelet coefficients and detects an arrival as a local maxima in a characteristic function.
Abstract: Typical microseismic data recorded by surface arrays are characterized by low signal-to-noise ratios (S/Ns) and highly nonstationary noise that make it difficult to detect small events. Currently, array or crosscorrelation-based approaches are used to enhance the S/N prior to processing. We have developed an alternative approach for S/N improvement and simultaneous detection of microseismic events. The proposed method is based on the synchrosqueezed continuous wavelet transform (SS-CWT) and custom thresholding of single-channel data. The SS-CWT allows for the adaptive filtering of time- and frequency-varying noise as well as offering an improvement in resolution over the conventional wavelet transform. Simultaneously, the algorithm incorporates a detection procedure that uses the thresholded wavelet coefficients and detects an arrival as a local maxima in a characteristic function. The algorithm was tested using a synthetic signal and field microseismic data, and our results have been compared wit...
216 citations
TL;DR: The proposed CSMcCulloch algorithm evolved to be most promising, and computationally efficient for segmenting satellite images, and outperforms others in attaining stable global optimum thresholds.
Abstract: This paper proposes a computationally efficient optimization algorithm for segmenting colour satellite images.CS algorithm incorporating Mantegna's and McCulloch's method for modeling levy flight is presented.PSO, DPSO, ABC and CS algorithms are compared with the proposed algorithm.All these optimization algorithms are exploited using three different objective functions.Performance assessment metrics demonstrated the improvement in the efficiency of the proposed algorithm. Satellite image segmentation is challenging due to the presence of weakly correlated and ambiguous multiple regions of interest. Several bio-inspired algorithms were developed to generate optimum threshold values for segmenting such images efficiently. Their exhaustive search nature makes them computationally expensive when extended to multilevel thresholding. In this paper, we propose a computationally efficient image segmentation algorithm, called CSMcCulloch, incorporating McCulloch's method for l e ? v y flight generation in Cuckoo Search (CS) algorithm. We have also investigated the impact of Mantegna's method for l e ? v y flight generation in CS algorithm (CSMantegna) by comparing it with the conventional CS algorithm which uses the simplified version of the same. CSMantegna algorithm resulted in improved segmentation quality with an expense of computational time. The performance of the proposed CSMcCulloch algorithm is compared with other bio-inspired algorithms such as Particle Swarm Optimization (PSO) algorithm, Darwinian Particle Swarm Optimization (DPSO) algorithm, Artificial Bee Colony (ABC) algorithm, Cuckoo Search (CS) algorithm and CSMantegna algorithm using Otsu's method, Kapur entropy and Tsallis entropy as objective functions. Experimental results were validated by measuring PSNR, MSE, FSIM and CPU running time for all the cases investigated. The proposed CSMcCulloch algorithm evolved to be most promising, and computationally efficient for segmenting satellite images. Convergence rate analysis also reveals that the proposed algorithm outperforms others in attaining stable global optimum thresholds. The experiments results encourages related researches in computer vision, remote sensing and image processing applications.
152 citations
TL;DR: This work proves that given data and a measurement matrix from a broad class of matrices, one can choose parameters for these classes of shrinkages to guarantee exact recovery of the sparsest solution.
Abstract: The minimization of compressed sensing is often relaxed to , which yields easy computation using the shrinkage mapping known as soft thresholding, and can be shown to recover the original solution under certain hypotheses. Recent work has derived a general class of shrinkages and associated nonconvex penalties that better approximate the original penalty and empirically can recover the original solution from fewer measurements. We specifically examine p-shrinkage and firm thresholding. In this work, we prove that given data and a measurement matrix from a broad class of matrices, one can choose parameters for these classes of shrinkages to guarantee exact recovery of the sparsest solution. We further prove convergence of the algorithm iterative p-shrinkage (IPS) for solving one such relaxed problem.
142 citations
TL;DR: In this article, a non-nondiagonal seismic denoising method based on the continuous wavelet transform with hybrid block thresholding (BT) was proposed for improving the signal-to-noise ratio of local microseismic, regional and ocean bottom seismic data.
Abstract: We introduce a nondiagonal seismic denoising method based on the continuous wavelet transform with hybrid block thresholding (BT). Parameters for the BT step are adaptively adjusted to the inferred signal property by minimizing the unbiased risk estimate of Stein (1980). The efficiency of the denoising for seismic data has been improved by adapting the wavelet thresholding and adding a preprocessing step based on a higher‐order statistical analysis and a postprocessing step based on Wiener filtering. Application of the proposed method on synthetic and real seismic data shows the effectiveness of the method for denoising and improving the signal‐to‐noise ratio of local microseismic, regional, and ocean bottom seismic data.
141 citations
01 Sep 2016
TL;DR: A new traffic sign detection and recognition method, which is achieved in three main steps, to use invariant geometric moments to classify shapes instead of machine learning algorithms and the results obtained are satisfactory when compared to the state-of-the-art methods.
Abstract: Graphical abstractDisplay Omitted In this paper we present a new traffic sign detection and recognition (TSDR) method, which is achieved in three main steps. The first step segments the image based on thresholding of HSI color space components. The second step detects traffic signs by processing the blobs extracted by the first step. The last one performs the recognition of the detected traffic signs. The main contributions of the paper are as follows. First, we propose, in the second step, to use invariant geometric moments to classify shapes instead of machine learning algorithms. Second, inspired by the existing features, new ones have been proposed for the recognition. The histogram of oriented gradients (HOG) features has been extended to the HSI color space and combined with the local self-similarity (LSS) features to get the descriptor we use in our algorithm. As a classifier, random forest and support vector machine (SVM) classifiers have been tested together with the new descriptor. The proposed method has been tested on both the German Traffic Sign Detection and Recognition Benchmark and the Swedish Traffic Signs Data sets. The results obtained are satisfactory when compared to the state-of-the-art methods.
TL;DR: The fully trained classifier provides an alternative to automated methods that require multi-modal MRI data, additional control scans, or user interaction to achieve optimal performance and has applications in neuroimaging and clinical contexts.
TL;DR: In this paper, a location-based thresholding approach based on sector inclusion with the Fama-French and SDPR sector Exchange Traded Funds (ETFs) is proposed.
Abstract: We document a striking block-diagonal pattern in the factor model residual covariances of the S&P 500 Equity Index constituents, after sorting the assets by their assigned Global Industry Classification Standard (GICS) codes. Cognizant of this structure, we propose combining a location-based thresholding approach based on sector inclusion with the Fama-French and SDPR sector Exchange Traded Funds (ETF’s). We investigate the performance of our estimators in an out-of-sample portfolio allocation study. We find that our simple and positive-definite covariance matrix estimator yields strong empirical results under a variety of factor models and thresholding schemes. Conversely, we find that the Fama-French factor model is only suitable for covariance estimation when used in conjunction with our proposed thresholding technique. Theoretically, we provide justification for the empirical results by jointly analyzing the in-fill and diverging dimension asymptotics.
TL;DR: Experimental results have shown that both SSO and FP algorithms outperform PSO and BAT algorithms while exhibiting equal performance for small numbers of thresholds, suggesting that the SSO algorithm can be effectively considered as an attractive alternative for the multilevel image thresholding problem.
Abstract: We present an empirical comparison of two new meta-heuristics SSO and FP.Real test images were used to perform thresholding using Otsu's method and Kapur's entropy.Compared algorithms were SSO, FP, PSO, BAT.Comparisons were made according to the fitness values, PSNR and SSIM.SSO shows superior performance in convergence and in quality terms. In this paper, we investigate the ability of two new nature-inspired metaheuristics namely the flower pollination (FP) and the social spiders optimization (SSO) algorithms to solve the image segmentation problem via multilevel thresholding. The FP algorithm is inspired from the biological process of flower pollination. It relies on two basic mechanisms to generate new solutions. The first one is the global pollination modeled in terms of a Levy distribution while the second one is the local pollination that is based on random selection of local solutions. For its part, the SSO algorithm mimics different natural cooperative behaviors of a spider colony. It considers male and female search agents subject to different evolutionary operators. In the two proposed algorithms, candidate solutions are firstly generated using the image histogram. Then, they are evolved according to the dynamics of their corresponding operators. During the optimization process, solutions are evaluated using the between-class variance or Kapur's method. The performance of each of the two proposed approaches has been assessed using a variety of benchmark images and compared against two other nature inspired algorithms from the literature namely PSO and BAT algorithms. Results have been analyzed both qualitatively and quantitatively based on the fitness values of obtained best solutions and two popular performance measures namely PSNR and SSIM indices as well. Experimental results have shown that both SSO and FP algorithms outperform PSO and BAT algorithms while exhibiting equal performance for small numbers of thresholds. For large numbers of thresholds, it was observed that the performance of FP algorithm decreases as it is often trapped in local minima. In contrary, the SSO algorithmprovides a good balance between exploration and exploitation and has shown to be the most efficient and the most stable for all images even with the increase of the threshold number. These promising results suggest that the SSO algorithm can be effectively considered as an attractive alternative for the multilevel image thresholding problem.
TL;DR: The results of the experimental work reveal that both approaches offer comparable or even better performances with respect to the best ones reported in the literature and are compatible to real-time operation as well.
Abstract: New methods are proposed for circular traffic sign detection and recognition.Comparable performances are attained with respect to the best performing methods.Compatibility to real-time operation is validated. Automatic traffic sign detection and recognition play crucial roles in several expert systems such as driver assistance and autonomous driving systems. In this work, novel approaches for circular traffic sign detection and recognition on color images are proposed. In traffic sign detection, a new approach, which utilizes a recently developed circle detection algorithm and an RGB-based color thresholding technique, is proposed. In traffic sign recognition, an ensemble of features including histogram of oriented gradients, local binary patterns and Gabor features are employed within a support vector machine classification framework. Performances of the proposed detection and recognition approaches are evaluated on German Traffic Sign Detection and Recognition Benchmark datasets, respectively. The results of the experimental work reveal that both approaches offer comparable or even better performances with respect to the best ones reported in the literature and are compatible to real-time operation as well.
TL;DR: In this article, a data-driven scheme for learning optimal thresholding functions for iterative shrinkage/thresholding algorithm (ISTA) is presented, which is obtained by relating iterations of ISTA to layers of a simple feedforward neural network and developing a corresponding error backpropagation algorithm for fine-tuning the thresholding function.
Abstract: Iterative shrinkage/thresholding algorithm (ISTA) is a well-studied method for finding sparse solutions to ill-posed inverse problems. In this letter, we present a data-driven scheme for learning optimal thresholding functions for ISTA. The proposed scheme is obtained by relating iterations of ISTA to layers of a simple feedforward neural network and developing a corresponding error backpropagation algorithm for fine-tuning the thresholding functions. Simulations on sparse statistical signals illustrate potential gains in estimation quality due to the proposed data adaptive ISTA.
TL;DR: A novel matched filter approach with the Gumbel probability distribution function as its kernel is introduced to improve the performance of retinal blood vessel segmentation and confirms that the proposed approach performance better with respect to other prominent Gaussian distribution function and Cauchy PDF based matched filter approaches.
TL;DR: A novel real time integrated method to locate the segmented region of interest of an image based on the Region Growing segmentation method along with the thresholding supported image segmentation established that the proposed integrated method outperformed the region growing method in terms of the recall and F-score.
Abstract: Image segmentation is a challenging process in numerous applications. Region growing is one of the segmentation techniques as a basis for the Seeded Region Growing method. A novel real time integrated method was developed in the current work to locate the segmented region of interest of an image based on the Region Growing segmentation method along with the thresholding supported image segmentation. Through the proposed work, a homogeneity based on pixel intensity was suggested as well as the threshold value can be decided via a variety of schemes such as manual selection, Iterative method, Otsu’s method, local thresholding to obtain the best possible threshold. The experimental results were performed on different images obtained from an Alpert dataset. A comparative study was arried out with the human segmented image, threshold based region growing, and the proposed integrated method. The results established that the proposed integrated method outperformed the region growing method in terms of the recall and F-score. Although, it had comparable recall values with that gained by the human segmented images. It was noted that as the image under test had a dark background with the brighter object, thus the proposed method provided the superior recall value compared to the other methods.
TL;DR: A novel hybrid differential evolution algorithm for selecting the optimal threshold values for a given gray-level input image, using the criterion defined by Otsu, which shows superior performance in the quality of the results.
Abstract: Real test images were used to perform thresholding using Otsu's method with a high level of thresholds.The proposed hybrid is compared with DE, jDE, PSO, ABC, and CS.Algorithms are compared based on PSNR and SSIM metrics.Friedman and Wilcoxon statistical tests are used to show the performances.Proposed hjDE shows superior performance in the quality of the results. Image thresholding is a process for separating interesting objects within an image from their background. An optimal threshold's selection can be regarded as a single objective optimization problem, where obtaining a solution can be computationally expensive and time-consuming, especially when the number of thresholds increases greatly. This paper proposes a novel hybrid differential evolution algorithm for selecting the optimal threshold values for a given gray-level input image, using the criterion defined by Otsu. The hybridization is done by adding a reset strategy, adopted from the Cuckoo Search, within the evolutionary loop of differential evolution. Additionally a study of different evolutionary or swarm-based intelligence algorithms for the purpose of thresholding, with a higher number of thresholds was performed, since many real-world applications require more than just a few thresholds for further processing. Experiments were performed on eleven real world images. The efficiency of the hybrid was compared to the cuckoo search and self-adaptive differential evolution, the original differential evolution, particle swarm optimization, and artificial bee colony where the results showed the superiority of the hybrid in terms of better segmentation results with the increased number of thresholds. Since the proposed method needs only two parameters adjusted, it is by far a better choice for real-life applications.
TL;DR: Experimental results reveal that the proposed defect detection model is effective and robust, and is superior than four existing models in terms of the high detection rate and low false alarm rate.
TL;DR: According to statistical analysis of different nature inspired optimization algorithms, Kapur's entropy was found to be more accurate and robust for multilevel colored satellite image segmentation problem, and cuckoo search was finding to be most promising for colored satellite images segmentation.
Abstract: This paper introduces the comparative performance of different objective functions (Kapur's & Otsu). Evolutionary algorithm based multilevel thresholding for a color satellite image has been presented. DE, WDO, PSO and CS algorithms are exploited with Kapur's and Otsu method. CS based Kapur's entropy was found to be more accurate for colored satellite image segmentation. Multilevel thresholding for segmentation is an essential task and indispensable process in various applications. Conventional color multilevel thresholding based image segmentations are computationally expensive, and lack accuracy and stability. To address this issue, this paper introduces the comparative performance study of different objective functions using cuckoo search and other optimization algorithms to solve the color image segmentation problem via multilevel thresholding. During the optimization process, solutions are evaluated using Otsu or Kapur's method. Performance of the proposed approach has been assessed using a variety of benchmark images, and compared against three other nature inspired algorithms namely differential evolution (DE), wind driven optimization (WDO) and particle swam optimization (PSO) algorithms. Results have been analyzed both qualitatively and quantitatively, based on the fitness values of obtained best solutions and four popular performance measures namely PSNR, MSE, SSIM and FSIM indices as well. According to statistical analysis of different nature inspired optimization algorithms, Kapur's entropy was found to be more accurate and robust for multilevel colored satellite image segmentation problem. On the other hand, cuckoo search was found to be most promising for colored satellite image segmentation.
01 Oct 2016
TL;DR: The experimental results demonstrate that the proposed CS-Kapur's energy curve based segmentation can powerfully and accurately search the multilevel thresholds.
Abstract: Display OmittedA complete flowchart routine of energy curve based multilevel image thresholding. Energy curve based colour multilevel thresholding has been proposed.CS and ELR-CS based optimization techniques have been exploited.Different objective functions have been utilized for optimum results.CS-Kapur's found giving better results. Amongst all the multilevel thresholding techniques, standard histogram based thresholding approaches are very impressive for bi-level thresholding. But, it is not effective to select spatial contextual information of the image for choosing optimal thresholds. In this paper, a new color image thresholding technique is presented by using an energy function to generate the energy curve of an image by considering spatial contextual information of the image. The property of this energy curve is very much similar to histogram of the image. To estimate the spatial contextual information for thresholding practice, in place of histogram, the energy curve function is used as an input. A new energy curve based color image segmentation approach using three well known objective functions named Kapur's entropy, between-class-variance, and Tsalli's entropy is proposed. In this paper, cuckoo search (CS) and egg lying radius-cuckoo search (ELR-CS) optimization algorithms with different parameter analysis have been used for solving the color image multilevel thresholding problem. The experimental results demonstrate that the proposed CS-Kapur's energy curve based segmentation can powerfully and accurately search the multilevel thresholds.
TL;DR: Experimental results show that the proposed method can detect the liver accurately with significant reduction of processing time, especially when dealing with diseased liver cases.
Abstract: Accurate segmentation of liver from abdominal CT scans is critical for computer-assisted diagnosis and therapy. Despite many years of research, automatic liver segmentation remains a challenging task. In this paper, a novel method was proposed for automatic delineation of liver on CT volume images using supervoxel-based graph cuts. To extract the liver volume of interest (VOI), the region of abdomen was firstly determined based on maximum intensity projection (MIP) and thresholding methods. Then, the patient-specific liver VOI was extracted from the region of abdomen by using a histogram-based adaptive thresholding method and morphological operations. The supervoxels of the liver VOI were generated using the simple linear iterative clustering (SLIC) method. The foreground/background seeds for graph cuts were generated on the largest liver slice, and the graph cuts algorithm was applied to the VOI supervoxels. Thirty abdominal CT images were used to evaluate the accuracy and efficiency of the proposed algorithm. Experimental results show that the proposed method can detect the liver accurately with significant reduction of processing time, especially when dealing with diseased liver cases.
01 Feb 2016
TL;DR: This paper proposes a video-based method for vehicle detection and counting system based on computer vision technology that uses background subtraction technique to find foreground objects in a video sequence.
Abstract: A vehicle detection and counting system plays an important role in an intelligent transportation system, especially for traffic management. This paper proposes a video-based method for vehicle detection and counting system based on computer vision technology. The proposed method uses background subtraction technique to find foreground objects in a video sequence. In order to detect moving vehicles more accurately, several computer vision techniques, including thresholding, hole filling and adaptive morphology operations, are then applied. Finally, vehicle counting is done by using a virtual detection zone. Experimental results show that the accuracy of the proposed vehicle counting system is around 96%.
TL;DR: The use of a novel iterative seislet-frame thresholding approach to separate the blended data in a distance-separated simultaneous-source acquisition with two sources has the potential to obtain better separated components with less artifacts.
Abstract: The distance-separated simultaneous-sourcing (DSSS) technique can make the smallest interference between different sources. In a distance-separated simultaneous-source acquisition with two sources, we propose the use of a novel iterative seislet-frame thresholding approach to separate the blended data. Because the separation is implemented in common shot gathers, there is no need for the random scheduling that is used in conventional simultaneous-source acquisition, where random scheduling is applied to ensure the incoherent property of blending noise in common midpoint, common receiver, or common offset gathers. Thus, DSSS becomes more flexible. The separation is based on the assumption that the local dips of the data from different sources are different. We can use the plane-wave destruction algorithm to simultaneously estimate the conflicting dips and then use seislet frames with two corresponding local dips to sparsify each signal component. Then, the different signal components can be easily separated. Compared with the FK-based approach, the proposed seislet-frame-based approach has the potential to obtain better separated components with less artifacts because the seislet frames are local transforms while the Fourier transform is a global transform. Both simulated synthetic and field data examples show very successful performance of the proposed approach.
TL;DR: Experimental results indicate that the proposed denoising method outperforms other denoised methodologies in terms of signal-to-noise ratio, mean square error, and percent root mean square difference.
Proceedings Article•
01 Apr 2016TL;DR: In this paper, a neural network is used to learn iterative sparse estimation algorithms for stereo estimation, where the goal is to remove sparse outliers that can disrupt the estimation of surface normals from a 3D scene.
Abstract: The iterations of many sparse estimation algorithms are comprised of a fixed linear filter cascaded with a thresholding nonlinearity, which collectively resemble a typical neural network layer. Consequently, a lengthy sequence of algorithm iterations can be viewed as a deep network with shared, hand-crafted layer weights. It is therefore quite natural to examine the degree to which a learned network model might act as a viable surrogate for traditional sparse estimation in domains where ample training data is available. While the possibility of a reduced computational budget is readily apparent when a ceiling is imposed on the number of layers, our work primarily focuses on estimation accuracy. In particular, it is well-known that when a signal dictionary has coherent columns, as quantified by a large RIP constant, then most tractable iterative algorithms are unable to find maximally sparse representations. In contrast, we demonstrate both theoretically and empirically the potential for a trained deep network to recover minimal $\ell_0$-norm representations in regimes where existing methods fail. The resulting system, which can effectively learn novel iterative sparse estimation algorithms, is deployed on a practical photometric stereo estimation problem, where the goal is to remove sparse outliers that can disrupt the estimation of surface normals from a 3D scene.
TL;DR: The discrete wavelet transform is employed to remove noise components of the time - frequency domain in order to enhance the ECG signal and the Hilbert transform with the adaptive thresholding technique used to explore an optimal combination to detect R-peaks more accurately.
TL;DR: In this paper, a modular and cost-effective AOI system for hot-rolled flat steel in real time is presented, where a defect detection algorithm is developed based on variance, entropy and average gradient derived from non-overlapping 32×32 pixel blocks of steel surface images.
Abstract: Defective steel brings economic and commercial reputation losses to the hot-strip manufacturers, and one of the main difficulties in using machine-vision-based technique for steel surface inspection is time taken to process the massive images suffering from uneven illumination. This paper develops a modular and cost-effective AOI system for hot-rolled flat steel in real time. Firstly, a detailed system topology is constructed according to the design goals covering the vast majority of steel mills, lighting setup and typical defect patterns are presented as well. Secondly, the image enhancement method is designed to overcome the uneven-lighting, over- or under-exposure. Thirdly, the defect detection algorithm is developed based on variance, entropy and average gradient derived from non-overlapping 32×32 pixel blocks of steel surface images. Fourthly, the proposed algorithms are implemented on FPGA in parallel to improve the inspection speed. Finally, 18,071 contiguous images (4096×1024 pixel) acquired from 7 defective steel rolls have been inspected by the realized AOI system to evaluate the performance. The experimental results show that the proposed method is speedy and effective enough for real applications in the hot-rolled steel manufacturing, with 92.11% average accuracy while 5.54% false-negative rate. HighlightsDetailed system topology is constructed according to the design goals covering the vast majority of steel mills.A new dynamic image enhancement method is designed to overcome the uneven-lighting, over- or under-exposure.A defect detection algorithm with adaptive thresholding mechanism is developed based on image block variance, entropy and average gradient.The inspection speed of the proposed algorithms running on FPGA is evaluated.Testing procedures and evaluation results from the applied hot-rolling mill are discussed.
TL;DR: A novel approach for unsupervised classification of land cover study of hyper-spectral satellite images to improve separation between objects and background by using multi-level thresholding based on the maximum Renyi entropy (MRE).
Abstract: Unsupervised classification of land cover study of hyper-spectral satellite images.A multi-level Renyi entropy based image thresholding scheme is presented.Multi-level thresholding is formulated as optimization problem and solved with DE.Composite kernel based classification approach using Support Vector Machine (SVM).Very competitive performance on popular hyper-spectral imagery like ROSIS and AVRIS. This article presents a novel approach for unsupervised classification of land cover study of hyper-spectral satellite images to improve separation between objects and background by using multi-level thresholding based on the maximum Renyi entropy (MRE). Multi-level thresholding, which partitions a gray-level image into several distinct homogeneous regions, is a widely popular tool for segmentation. However, utility of multi-level thresholding is yet to be investigated in challenging applications like hyper-spectral image analysis. Differential Evolution (DE), a simple yet efficient evolutionary algorithm of current interest, is employed to improve the computation time and robustness of the proposed algorithm. The performance of DE is also investigated extensively through comparison with other well-known nature inspired global optimization techniques. In addition, the outcomes of the MRE-based thresholding are employed to train a Support Vector Machine (SVM) classifier via the composite kernel approach to improve the classification accuracy. The final outcomes are tested on popular hyper-spectral imagery like ROSIS and AVRIS sensors. The effectiveness of the proposed algorithm is evaluated through qualitative and quantitative comparison with other state-of-the-art global optimizers.