Distance Transform Research Articles

Speaker verification normalization sequence kernel based on Gaussian mixture model super-vector and Bhattacharyya distance

Due to the accessibility and economy of human speech, speaker verification has become the research hotspot in the field of biometric authentication. A novel normalization sequence kernel based on Bhattacharyya distance clustering and within class covariance normalization was proposed in this paper. In this kernel, the high computation complexity and channel interference susceptibility that commonly exist in speaker verification could be restrained. In our method, we calculated the Bhattacharyya distance between pairs of Gaussian mixture models first. And then, a clustering algorithm was designed according to Gaussian mixture model’s Bhattacharyya distance to obtain clustering center models. Maximum a posteriori was applied on these clustering center models to generate super-vectors immediately following. The sequence kernel was generated based on Bhattacharyya distance transformation and super-vectors. Finally, within class covariance normalization was utilized to restrain the channel distortion in kernel space. We adopted support vector machine as classifier to decide the target speaker. The experiment results on TIMIT corpus and NIST 2008 SRE showed that our proposed kernel has superior recognition accuracy and better robustness.

A novel voxel-based method to estimate cortical sulci width and its application to compare patients with Alzheimer’s disease to controls

A voxel-based method for measuring sulcal width was developed, validated and applied to a database. This method (EDT-based LM) employs the 3D Euclidean Distance Transform (EDT) of the pial surface and a Local Maxima labeling algorithm. A computational phantom was designed to test method performance; results revealed the method’s inaccuracy δ, to range between 0.1 and 0.5 voxels, for a width that varied between 1 and 7 voxels. Two morphological descriptors were computed to characterize each defined sulcus: mean sulcal width (MSW) and mean absolute deviation (MAD). The former is the average width for all available width measurements within the sulcus, and the latter is the deviation of these measurements. The EDT-based LM method was applied to the Minimal Interval Resonance Imaging in the Alzheimer’s Disease (MIRIAD) database, for a set of high-resolution Magnetic Resonance (MR) images of 66 subjects: 43 patients with Alzheimer Disease (AD) and 23 control subjects. AD causes significant gray matter loss; hence, some sulci were expected to broaden. Methodological results concurred with this hypothesis. After a Wilcoxon test, MSW was grater in the case of all sulci pertaining to AD patients, (p < 0.05, FDR corrected), whereas MAD showed significant differences in 8 sulci (p < 0.05, FDR corrected). This work presents a novel voxel-based method for measuring sulcal width and extracting descriptors to characterize and compare the sulci within and across subjects.

A Model for Animal Home Range Estimation Based on the Active Learning Method

Home range estimation is the basis of ecology and animal behavior research. Some popular estimators have been presented; however, they have not fully considered the impacts of terrain and obstacles. To address this defect, a novel estimator named the density-based fuzzy home range estimator (DFHRE) is proposed in this study, based on the active learning method (ALM). The Euclidean distance is replaced by the cost distance-induced geodesic distance transformation to account for the effects of terrain and obstacles. Three datasets are used to verify the proposed method, and comparisons with the kernel density-based estimator (KDE) and the local convex hulls (LoCoH) estimators and the cross validation test indicate that the proposed estimator outperforms the KDE and the LoCoH estimators.

Fractal analysis and spatial disposition of porosity in major indian gas shales using low-pressure nitrogen adsorption and advanced image segmentation

Pore surface heterogeneities and their associated lacunarity of major Indian shale gas reservoirs were analyzed using both gas adsorption and advanced image analysis. The experiments were carried on samples from priority Indian gas shales namely Cambay, Cauvery, Krishna-Godavari and Damodar Valley. An advanced watershed segmentation was adopted on the SEM images to understand the control of pore size on the porosity of shale formations. Further, 3D as well as 2D surface intensity maps were plotted from the SEM images applying the distance transformation technique. The fractal dimension D ranged from 2.55 to 2.78 for LPN2GA suggesting a high degree of complexity in their fractal pattern. The 3D surface intensity map is found to be useful in picturizing the anisotropy in shale brought about by preferential orientation of clay minerals. The results can be extended in understanding the behavior of gas flow within the complex pore systems as exhibited by shale.

Distance transform and template matching based methods for localization of barcodes and QR codes

Visual codes play an important role in automatic identification, which became an inseparable part of industrial processes. Thanks to the revolution of smartphones and telecommunication, it also becomes more and more popular in everyday life, containing embedded web addresses or other small informative texts. While barcode reading is straightforward in images having optimal parameters (focus, illumination, code orientation, and position), localization of code regions is still challenging in many scenarios. Every setup has its own characteristics, therefore many approaches are justifiable. Industrial applications are likely to have more fixed parameters like illumination, camera type and code size, and processing speed and accuracy are the most important requirements. In everyday use, like with smartphone cameras, a wide variety of code types, sizes, noise levels and blurring can be observed, but the processing speed is often not crucial, and the image acquisition process can be repeated in order for successful detection. In this paper, we address this problem with two novel methods for localization of 1D barcodes based on template matching and distance transformation, and a third method to detect QR codes. Our proposed approaches can simultaneously localize several different types of codes. We compare the effectiveness of the proposed methods with several approaches from the literature using public databases and a large set of synthetic images as a benchmark. The evaluation shows that the proposed methods are efficient, having 84.3% Jaccard accuracy, superior to other approaches. One of the presented approaches is an improvement on our previous work. Our template matching based method is computationally more complex, however, it can be adapted to specific code types providing high accuracy. The other method uses distance transformation, which is fast and gives rough regions of interests that can contain valid visual code candidates.

New Accuracy Evaluation Index for Track Fusion Algorithms

When evaluating the track fusion algorithm, common accuracy indexes may fail to evaluate the fusion accuracy correctly when the state estimation and the real target cannot be one-to-one, and fail to effectively distinguish the performance of the algorithm when the state estimation is similar. Therefore, it is necessary to construct a high-resolution evaluation index, which can evaluate the track fusion algorithm more accurately, reasonably and comprehensively. Firstly, the advantages and disadvantages of the optimal subpattern assignment (OSPA) distance as the accuracy index to evaluate the track fusion algorithm are analyzed. Then, its deficiencies are improved by using the Hellinger distance instead of the original Euclidean distance, and the distance is index transformed. Finally, a new evaluation index for track fusion algorithms is proposed, which is the OSPA distance based on Hellinger distance and index transformation. The simulation results show that the new index can not only correctly evaluate the fusion precision, but also consider the state uncertainty, making that can evaluate the track fusion algorithm more sensitively, and effectively solves the sensitivity of the index to the cut-off parameter c through index transformation.

The Distance-Sinh combined transformation for near-singularity cancelation based on the generalized Duffy normalization

The singularity cancelation is usually of great importance in the calculation of nearly singular integral that is involved in the boundary element analysis of thin structures and in the extended finite element method for crack problems. The well-known distance transformation can be performed for various kinds of singular kernels to compute the near singular integral. However, in the case of that the source point is located nearby the boundary of the integral element, the result is usually very sensitive to the shape of the sub-triangles. A Distance-Sinh combined transformation method based on the generalized Duffy normalization is proposed in this paper to improve the accuracy of the nearly singular integral calculation. Furthermore, a two-level transformation method is proposed to cancel the singularity of the integrand. In first level of the method, the generalized Duffy transformation is implemented. In the transformed integral space (u, v), the singularity of the integrand along u-direction is eliminated by the distance transformation. The singularity of the integrand along the other direction is canceled by the Sinh transformation. Numerical examples are presented to verify the proposed method.

Shape description and recognition by implicit Chebyshev moments

Chebyshev Moments(CMs) have been applied to representation and recognition of 2D object shapes in image processing and computer vision. However they still suffer from poor representation power and difficulty in computing invariants for shapes. In this work, we present Implicit Chebyshev Moments (ICMs) to overcome these issues. Firstly, we use Euclid distance transformation to generate a series of level sets based on a given shape. Secondly, we fit an implicit Chebyshev polynomial to the data set consisting of the obtained level sets together with all the boundary points on the original shape and call the obtained coefficients of the fitted implicit Chebyshev polynomial ICMs. Finally, we propose a new approach to derive geometric invariants based on ICMs. In addition, we also develop an algorithm for the determination of a suitable degree for implicit Chebyshev polynomials before representing a given shape. Experimental results show the ICMs are more efficient for representing complex shapes than CMs.

A Face Alignment Accelerator Based on Optimized Coarse-to-Fine Shape Searching

The coarse-to-fine shape searching (CFSS) framework is a recently developed algorithm that achieves relatively high accuracy in face alignment by alleviating the poor initialization problem facing traditional cascaded regression approaches. However, its high computational complexity and memory access demands make it difficult for CFSS to satisfy the requirements of real-time processing. To address this issue, a fast shape searching face alignment (F-SSFA) accelerator is presented based on the optimization of the CFSS algorithm and an efficient hardware implementation. First, the learning-based low-dimensional speeded-up robust features method, based on the correlations between the SURF features and the regression targets, is introduced to distill the feature set down to the only most distinct features to reduce the computing load. Second, the partial keypoints Euclidean distance and shape affine transformation are introduced to replace feature extraction and support vector machine classification, thereby accelerating the shape searching process. Compared with CFSS, F-SSFA achieves a $5.8\times $ speedup while achieving similar accuracy. Moreover, a VLSI architecture is proposed to realize the fixed-point F-SSFA algorithm. Multiple descriptors located in adjacent regions are simultaneously generated in a single access to the corresponding image data. Therefore, repeated memory access operations are avoided. The optimal parameter configuration for hardware implementation is also exploited based on a tradeoff between accuracy and hardware performance. Simulated with TSMC 65-nm 1P8M technology within a 3.6 mm2 area, a post-layout simulation shows that 700 fps can be achieved while consuming 300 mW at 200 MHz.

A Mid-Axis extraction of heterogeneous river network data with topological consistency

Abstract. Geographic data in the big data environment presents some new characteristics: diversified data sources, and different ways of obtaining data, which makes the form and content of data different, resulting in diversification of GIS data types. Multiple sources and multiple scales characterize line-surface mixed river data. Different acquisition methods, different acquisition times, and different scales will result in differences in line-surface data and even different resolutions. The river network data used in this experiment usually includes 3 types of river data: surface (Dual-line River), single-line rivers and raster river data from remote sensing images. Among them, the river raster data can be treated as surface data. The surface river data has some specific characteristics. First, the abutment relationship between the line and the surface such as the tributary flow often exists on the main stream. The planar rivers often connect with the linear rivers. Second, the inclusion relationship between the surface and the surface, such as the surface Island phenomenon in the river. Last, the relationship between the two lines and the line, such as the adjoining, intersecting, and separating of the two single-line rivers.The following methods are used to achieve network topology with line-surface river data. First, the line-surface river data was rasterized into a unified resolution raster so that they have the same data characteristics. Second, the Mid-Axis of the line-surface data is extracted through distance transformation or refine method from mathematical morphology. At last we can establish a topological relationship with the linear data, realize the network topology and apply to network analysis.The blue line in Fig. 1a is based on the linear extraction result of the line-surface mixed data of the Delaunay triangulation, and the line in Fig. 1b is the central axis extraction result of the proposed algorithm. Focus on the connection of the line surface. As can be seen from (a), for the mixed data Delaunay triangulation, only the axis of the planar element can be extracted, and the original single line data cannot be retained. The central axis and the original line data are not established. Topological relationship. The morphological-based refinement method proposed in Figure 1(b) can well avoid the above problems.For the real data experiment, the single-line river that was originally rasterized by linear rivers is well preserved. When a linear river overlaps or is connected to a planar river. The extracted central axis and the single-line river can remain in a completely connected state whenever overlap or junction, and will not adversely affect the extraction result, realizing a fully automated central axis extraction.The method proposed in this paper is characterized by satisfying the basic requirements of the visualization of the central axis extraction result and achieving full automatic extraction. After the topology check of the extraction result is completed, the network space analysis is directly performed. For example, for the river network axis can be connected for connectivity analysis. Since the data selected in this paper is river network data, the following will establish a network topology and perform connectivity analysis on the results of the central axis extraction.The two points in Figure 3 are the starting point, point 1 is on the axis of the original linear river, and point 2 is on the central axis of the planar river. The connectivity path is generated between the two points as shown in the figure. The results show that the single-line river network composed of the axes in the line-mixed data extraction has a complete topological connectivity.River networks are complex spatial data with highly structured features with different scales. The acquisition and establishment of single-line topology river network data is essential for map synthesis and solving the topological relationship of watershed systems, and will play a huge role in GIS applications.Based on the mathematical morphology and Map Algebra principle, this paper uses the refinement algorithm and template shaping to realize the extraction of the multi-source and multi-scale river central axes of the line-surface mixing, and it is excellent in the transition from the two-line river to the single-line river. The test results show that even for the mixed data of the line surface (including islands) with complex structure, the method can effectively extract the continuous and smooth central axis and maintain a complete topological relationship with the retained single line data. The acquisition and establishment of single-line topology river network data is realized. In the next step of the work, the algorithm will be optimized to improve the speed of the algorithm. Although the study in this paper is based on rivers, the research results are equally applicable to other multi-source mixed line data, such as street networks.

A Work Efficient Parallel Algorithm for Exact Euclidean Distance Transform.

A fully-parallelized work-time optimal algorithm is presented for computing the exact Euclidean Distance Transform (EDT) of a 2D binary image with the size of n×n . Unlike existing PRAM (Parallel Random Access Machine) and other algorithms, this algorithm is suitable for implementation on modern SIMD (Single Instruction Multiple Data) architectures such as GPUs. As a fundamental operation of 2D EDT, 1D EDT is efficiently parallelized first. Specifically, the GPU algorithm for the 1D EDT, which uses CUDA (Compute Unified Device Architecture) binary functions, such as ballot(), ffs(), clz(), and shfl(), runs in O(log32n) time and performs O(n) work. Using the 1D EDT as a fundamental operation, the fully-parallelized work-time optimal 2D EDT algorithm is designed. This algorithm consists of three steps. Step 1 of the algorithm runs in O(log32n) time and performs O(N) ( N = n2 ) of total work on GPU. Step 2 performs O(N) of total work and has an expected time complexity of O(logn) on GPU. Step 3 runs in O(log32n) time and performs O(N) of total work on GPU. As far as we know, this algorithm is the first fully-parallelized and realized work-time optimal algorithm for GPUs. The experimental results show that this algorithm outperforms the prior state-of-the-art GPU algorithms.

Dense networks with relative location awareness for thorax disease identification.

Chest X-ray is one of the most common examinations for diagnosing heart and lung diseases. Due to the existing of a large number of clinical cases, many automated diagnosis algorithms based on chest X-ray images have been proposed. To our knowledge, almost none of the previous auto-diagnosis algorithms consider the effect of relative location information on disease incidence. In this study, we propose to use relative location information to assist the identification of thorax diseases. In this work, U-Net is used to segment lung and heart from chest image. The relative location maps are computed through Euclidean distance transformation from segmented masks. By introducing the relative location information into the network, the usual location of disease is combined with the incidence. The proposed network is the fusion of two branches: mask branch and image branch. A mask branch is designed to be a bottom-up and top-down structure to extract relative location information. The structure has a large receptive field, which can extract more information for large lesion and contextual information for small lesion. The features learned from mask branch are fused with image branch, which is a 121-layers DenseNet. We compare our proposed method with four state-of-the-art methods on the largest public chest X-ray dataset: ChestX-ray14. The proposed method achieves the area under a curve of 0.820, which outperforms all the existing models and algorithms. This paper proposed a dense network with relative location information to identify thorax disease. The method combines the usual location of disease with the incidence for the first time and performs good.

Adaptive and Robust Multiple Image Watermarking Using Canberra Distance and Dual Tree Complex Wavelet Transform

Adaptive and Robust Multiple Image Watermarking Using Canberra Distance and Dual Tree Complex Wavelet Transform

Novel microcomputed tomography method for thickness analysis of calcified cartilage

Purpose: Understanding of calcified cartilage morphology could provide new cues on osteoarthritis development. However, since calcified cartilage and subchondral bone share similar X-ray attenuation properties their separation from micro-computed tomography (micro-CT) images remains challenging. This study aimed at developing and validating a novel method for the segmentation and thickness analysis of calcified cartilage (CC) using micro-CT. Methods: To facilitate high-resolution micro-CT imaging, eight healthy knees from skeletally mature rabbits (N = 8) were dissected to six smaller osteochondral samples, i.e. lateral and medial femoral condyles, femoral groove, lateral and medial tibial plateau as well as patella (in total n = 48 samples). The samples were imaged with micro-CT device (isotropic voxel size: 3.2 μm). After imaging, the samples were decalcified and embedded in paraffin. Three histology sections (∼5 μm thickness) were stained with Goldner’s trichrome from each sample. Preprocessing for CC segmentation: The micro-CT data were reconstructed into two image stacks with: 1) standard and 2) mineral windows, respectively. The standard window (SW) image stack contained information from a wide attenuation coefficient range, while the mineral window (MW) image stack contained information from a narrow attenuation coefficient range providing a high contrast between CC and bone. Both stacks were co-registered. Trabecular bone volume-of-interest (VOItrab) was obtained from the SW stack with manual segmentation. VOItrab was subtracted from the SW stack. The output stack was inverted and refined with a shrink-wrap operation. This generated a subchondral plate VOI (VOIplate) including both subchondral bone and CC. CC segmentation: The region defined by VOIplate was selected from the MW stack and a narrow, high threshold window was applied to select only subchondral bone. Subsequently, the selected subchondral bone was subtracted from the VOIplate region to produce the CC segmentation. Finally, morphological opening operations were used for eliminating the edge artifact for more realistic estimation of CC thickness (CC.Th). CC thickness analysis: First, 2D analysis both from micro-CT and histology was conducted to validate the micro-CT-based CC.Th analysis. The CC.Th obtained from micro-CT (as an average of five slices) was compared with the CC.Th from histology sections using ImageJ (BoneJ plugin). Then, CC.Th was analyzed in 3D using a sphere-based method utilizing a distance transformation. Statistics: Normality of the CC.Th data were first ensured (Shapiro-Wilk test). The agreement between the 2D micro-CT and histology analysis was evaluated with a Bland-Altman plot. One-way ANOVA test (post-hoc test: Newman-Keuls) was used for comparing the CC.Th between different locations. The level of significance was set to 0.05. Results: CC.Th in 2D had a good agreement between histology and micro-CT (Figure 1A to 1C, bias: -8.6 μm, levels of agreement: [-71.3, 54.1] μm). In the 2D histology analysis (Figure 1D), the patella had the thickest CC, while the thinnest locations were found for the tibial plateau and medial femoral condyle. Interestingly, in all 2D and 3D analyses (Figure 1E and 1F) CC was always thicker in the patella compared to the femoral groove and tibial plateau. Furthermore, CC in the lateral femoral condyle was always thicker than in the medial tibial plateau. Conclusions: In this study, we presented and validated a novel method for the thickness analysis of CC from micro-CT images. In addition, we observed different CC.Th. between anatomical locations with both histology and micro-CT. The different CC.Th. between anatomical locations is most likely related with local loading conditions.In the future, this method can be used for assessing the potential changes of CC.Th. for different joint pathologies, such as osteoarthritis.

High order isoparametric elements in boundary element method—Smooth spheroidal element

The smooth spheroidal element, a high order isoparametric element, is constructed and applied in the boundary element method (BEM) for modeling closed surface of spheroid as well as the field variables using a single element. The smoothness of the element is realized by repeated use of real nodes as auxiliary nodes of interpolation along both the latitude and meridian directions to remove the end node/line effects existing in the conventional closure element. The special techniques of shape function manipulations are suggested and presented to deal with the integrals with hyper- or near hyper-singularities in a uniform and simple way encountered in the BEM. A combined technique, the distance transformation with possible subdivision is proposed to improve the evaluation of near singular integrals. Through a number of numerical examples, the accuracy and efficiency of using high order smooth elements are demonstrated by checking the fitting accuracy of geometrical parameters, the computation costs and some bench mark tests in elasticity.

Classification for plastic bottles recycling based on image recognition

Recycling of used plastic bottles is an important measure to protect the environment and save energy. Usually, bottles in different colors have different value for recycling. Classification of plastic bottles recycling based on image recognition during recycling is an effective way, where the position and color recognition are the key technologies. To classify the plastic bottles on the conveyor belt, their position relationships are firstly defined as three categories, i.e. disjoint, adjacent and overlapping. The disjoint ones can be easily identified by the ratio of concave and convex area based on their image. For the adjacent and overlapping bottles, a combination method called distance transformation and threshold segmentation is proposed to distinguish their position relationships. Once the adjacent bottles are identified, the method of concave point search based on convex hull will be used to separate the adjacent recycled bottles further. Then, the color of both the disjoint and adjacent bottles is identified because it is too complex and difficult to recognize color of and separate the overlapping bottles. In the aspect of color recognition, the colors of recycled bottles are divided into seven categories in the sorting process. Color features of the bottom section are used to represent the one of the recycled bottle because there may be a bottle cap and a label on the top and in the middle of the bottle, respectively, resulting in the wrong recognition. ReliefF algorithm is applied to select color features of recycled bottles and the color is identified by support vector machine (SVM) algorithm. The influence of training sample size on classification model is studied and the experimental results show that the accuracy of color recognition of recycled bottles reach 94.7%.

Generation of Thangka relief from line drawings

Line drawings lack the three-dimensional (3D) information of the depicted objects, which is present in 3D models and photographs based on relief-generation techniques. Therefore, creating relief surfaces from line drawings presents special research challenges. This paper creates a 3D relief model of Thangka, the Intangible Cultural Heritage of China and UNESCO, from line drawings. To obtain the relief surfaces, our approach solves Poisson equations under the constraints of gray-scale control pictures and boundary seeds of the object regions. The Thangka relief is divided into low-frequency, mid-frequency and high-frequency components. The low- and mid-frequency components respectively correspond to the height of the human body and the limbs in the relief model, and are obtained by solving the Poisson equations constrained by distance transformations. The high-frequency components correspond to the heights of local details such as eyebrows, flowers, clouds, and mountains in the relief model, and are obtained by solving the Poisson equations with boundary constraints. Finally we add the height maps of the three components and transform the resultant height map into a mesh model of the relief. The effectiveness of our approach is demonstrated in several examples.

Canny-VO: Visual Odometry With RGB-D Cameras Based on Geometric 3-D–2-D Edge Alignment

This paper reviews the classical problem of free-form curve registration and applies it to an efficient RGB-D visual odometry system called Canny-VO, as it efficiently tracks all Canny edge features extracted from the images. Two replacements for the distance transformation commonly used in edge registration are proposed: approximate nearest neighbor fields and oriented nearest neighbor fields. 3-D-2-D edge alignment benefits from these alternative formulations in terms of both efficiency and accuracy. It removes the need for the more computationally demanding paradigms of data-to-model registration, bilinear interpolation, and subgradient computation. To ensure robustness of the system in the presence of outliers and sensor noise, the registration is formulated as a maximum a posteriori problem and the resulting weighted least-squares objective is solved by the iteratively reweighted least-squares method. A variety of robust weight functions are investigated and the optimal choice is made based on the statistics of the residual errors. Efficiency is furthermore boosted by an adaptively sampled definition of the nearest neighbor fields. Extensive evaluations on public SLAM benchmark sequences demonstrate state-of-the-art performance and an advantage over classical Euclidean distance fields.

Text localization in camera captured images using fuzzy distance transform based adaptive stroke filter

Localization of text from camera captured images with complex background is now-a-days a growing demand of modern IT enable service. Most of the current text localization techniques are sensitive to text features like color, size, style and also to the background clutter. Among all the methods proposed in different literatures, Stroke Filter is much more effective in localization of text. The effectiveness of traditional stroke filter is limited because of its fixed width and is capable of segmenting strokes/texts of predefined range of width. The proposed method uses Fuzzy Distance Transform based adaptive stroke filter which can effectively localize text regions from camera captured images with complex background. The method is applied by experiment on a database containing 600 images and the visual response of text segmentation is quite impressive. To get the accuracy of the proposed method, it is applied on a set of 16 test images and the segmentation result is compared with the ground truth images resulting in a recall, precision and f-measure values of 96.65%, 87.77% and 91.89% respectively.

Automatic Seal Imprint Verification Systems Using Edge Difference

In Asian countries, seals are widely used for authenticating the identity of a person or an organization. Therefore, the ability to efficiently verify whether a seal is either genuine or forged is important. Both rectangular and circular seal imprints are considered in this work. Geometric transformation is performed to align the perspective of the detected imprint image with the reference imprint. After the edge-differences between the reference image and the detected image are computed, a distance transformation is applied to extract several kinds of weighting matrices. The edge-difference images are weighted element-wise with the edge-difference images to calculate some predefined similarity metrics which serve as the input to the SVM (support vector machine). The imprint is then determined to be either genuine or forged. The experimental results show the effectiveness of the proposed verification approach.