Local Shape Features Research Articles

Shape optimization of a flexible beam with a local shape feature based on ANCF

Shape optimization of a flexible beam with a local shape feature based on ANCF

Deep convolutional networks do not classify based on global object shape.

Deep convolutional networks (DCNNs) are achieving previously unseen performance in object classification, raising questions about whether DCNNs operate similarly to human vision. In biological vision, shape is arguably the most important cue for recognition. We tested the role of shape information in DCNNs trained to recognize objects. In Experiment 1, we presented a trained DCNN with object silhouettes that preserved overall shape but were filled with surface texture taken from other objects. Shape cues appeared to play some role in the classification of artifacts, but little or none for animals. In Experiments 2–4, DCNNs showed no ability to classify glass figurines or outlines but correctly classified some silhouettes. Aspects of these results led us to hypothesize that DCNNs do not distinguish object’s bounding contours from other edges, and that DCNNs access some local shape features, but not global shape. In Experiment 5, we tested this hypothesis with displays that preserved local features but disrupted global shape, and vice versa. With disrupted global shape, which reduced human accuracy to 28%, DCNNs gave the same classification labels as with ordinary shapes. Conversely, local contour changes eliminated accurate DCNN classification but caused no difficulty for human observers. These results provide evidence that DCNNs have access to some local shape information in the form of local edge relations, but they have no access to global object shapes.

ShapeGTB: the role of local DNA shape in prioritization of functional variants in human promoters with machine learning.

MotivationThe identification of functional sequence variations in regulatory DNA regions is one of the major challenges of modern genetics. Here, we report results of a combined multifactor analysis of properties characterizing functional sequence variants located in promoter regions of genes.ResultsWe demonstrate that GC-content of the local sequence fragments and local DNA shape features play significant role in prioritization of functional variants and outscore features related to histone modifications, transcription factors binding sites, or evolutionary conservation descriptors. Those observations allowed us to build specialized machine learning classifier identifying functional single nucleotide polymorphisms within promoter regions—ShapeGTB. We compared our method with more general tools predicting pathogenicity of all non-coding variants. ShapeGTB outperformed them by a wide margin (average precision 0.93 vs. 0.47–0.55). On the external validation set based on ClinVar database it displayed worse performance but was still competitive with other methods (average precision 0.47 vs. 0.23–0.42). Such results suggest unique characteristics of mutations located within promoter regions and are a promising signal for the development of more accurate variant prioritization tools in the future.

Parameter optimization criteria guided 3D point cloud classification

3D point cloud classification is one of the basic topics in multimedia analysis and understanding. By the construction of the discriminant model and efficient parameter optimization, point cloud classification can be achieved after the training. However, most parameter optimization methods do not guarantee the highest global classification accuracy with a high classification accuracy on smaller classes. In addition, geometric features of the point cloud are not sufficiently utilized. In this paper, we use local geometric shape features including the nearest neighbor tetrahedral volume, Gaussian curvature, the neighbourhood normal vector consistency and the neighbourhood minimum principal curvature direction consistency. We propose three discrete criteria for parameter optimization to design explicit functions, and we present concrete algorithms, in which Monte Carlo method and Probabilistic Neural Network method are employed to estimate these parameters respectively. Experimental results show that our criteria can be applied to the classification of the 3D point cloud of the scene, and can be used to improve the classification accuracy of small-scale point sets when different classes have great disparities in the number.

Occluded information is restored at preview but not during visual search.

Objects often appear with some amount of occlusion. We fill in missing information using local shape features even before attending to those objects—a process called amodal completion. Here we explore the possibility that knowledge about common realistic objects can be used to “restore” missing information even in cases where amodal completion is not expected. We systematically varied whether visual search targets were occluded or not, both at preview and in search displays. Button-press responses were longest when the preview was unoccluded and the target was occluded in the search display. This pattern is consistent with a target-verification process that uses the features visible at preview but does not restore missing information in the search display. However, visual search guidance was weakest whenever the target was occluded in the search display, regardless of whether it was occluded at preview. This pattern suggests that information missing during the preview was restored and used to guide search, thereby resulting in a feature mismatch and poor guidance. If this process were preattentive, as with amodal completion, we should have found roughly equivalent search guidance across all conditions because the target would always be unoccluded or restored, resulting in no mismatch. We conclude that realistic objects are restored behind occluders during search target preview, even in situations not prone to amodal completion, and this restoration does not occur preattentively during search.

A design for a dynamic biomimetic sonarhead inspired by horseshoe bats

The noseleaf and pinnae of horseshoe bats (Rhinolophus ferrumequinum) have both been shown to actively deform during biosonar operation. Since these baffle structures directly affect the properties of the animals biosonar system, this work mimics horseshoe bat sonar system with the goal of developing a platform to study the dynamic sensing principles horseshoe bats employ. Consequently, two robotic devices were developed to mimic the dynamic emission and reception characteristics of horseshoe bats. The noseleaf and pinnae shapes were modeled as smooth blanks matched to digital representations of a horseshoe bat specimens noseleaf and pinnae. Local shape features mimicking structures on the pinnae and noseleaf were added digitally. Flexible baffles with local shape feature combinations were manufactured and paired with actuation mechanisms to mimic pinnae and noseleaf deformations in vivo. Two noseleaves with and without local shape features were considered. Each noseleaf baffle was mounted to a platform called the dynamic emission head to actuate three surface elements of the baffle. Similarly, 12 pinna realizations composed of combinations of three local shape features were mounted to a platform called the dynamic reception head to deform the left and right pinnae independently. Motion of the noseleaf and pinnae were synchronized to the incoming and outgoing sonar waveform, and the joint time-frequency properties of the noseleaf and pinnae local feature combinations and pairs of pinnae and noseleaf thereof were characterized across spatial direction. Amplitude modulations to the outgoing and incoming sonar pulse information across spatial direction were observed for all pinnae and noseleaf local shape feature combinations. Peak modulation variance generated by motion of the pinnae and combinations of the noseleaf and pinnae approached a white Gaussian noise variance bound. It was found the dynamic emitter generated less modulation than either the combined or reception scenarios.

Visual features based boosted classification of weeds for real-time selective herbicide sprayer systems

Recent years have shown enthusiastic research interest in weed classification for selective herbicide sprayer systems which are helpful in eradicating unwanted plants such as weeds from fields, minimizing the side effects of chemicals on the environment and crops. Two commonly found weeds are monocots (thin leaf) and dicots (broad leaf), requiring separate chemical herbicides for eradication. Researchers have used various computer vision-assisted techniques for eradication of these weeds. However, the changing and un-predictive lighting conditions in fields make the process of weed detection and identification very challenging. Therefore, in this paper, we present an efficient weed classification framework for real-time selective herbicide sprayer systems, exploiting boosted visual features of images, containing weeds. The proposed method effectively represents the image using local shape and texture features which are extracted during the leaf growth stage using an efficient method, preserving the discrimination between various weed species. Such effective representation allows accurate recognition at early growth stages. Furthermore, the various illumination problems prior to feature extraction are minimized using an adaptive segmentation algorithm. AdaBoost with Naïve Bayes as a base classifier discriminates the two weed species. The proposed method achieves an overall accuracy 98.40%, with true positive rate of 0.983 and false positive rate of 0.0121 for the original dataset and achieved 94.72% accuracy with the expanded dataset. The execution time of the proposed method is about 35 millisecond per image, which is less than state-of-the-art methods.

Biosonar beampattern measurements in flying old world leaf-nosed bats

Emission beampatterns, i.e., emitted ultrasound amplitude as a function of direction and frequency, could be an important characteristic of a bat's biosonar since they determine which part of an environment is illuminated with a given frequency and at a given time. Numerical results have predicted intricate geometrical features such as multiple peaks and notches in bat biosonar beampatterns. Furthermore, it has been shown that bats with nasal emission change the shapes of their emission baffles, “noseleaves,” during pulse emission. The work presented here is aimed at understanding if and how these two aspects (beam geometry and emission dynamics) are integrated into the bats' biosonar behaviors on the wing. The experiments are conducted with Old World leaf-nosed bats (Hipposideros pratti and Hipposideros armiger) that have a pronounced noseleaf dynamics. The animals fly in a tunnel that is instrumented with synchronized arrays of high-speed video cameras and microphones. The setup allows for precise tracking of the bats with high spatial resolution as well as beampattern estimates as a function of time. A first pilot dataset has indicated the presence of local shape features in the beampattern that can also vary with time. However, further testing is needed to corroborate these findings.Emission beampatterns, i.e., emitted ultrasound amplitude as a function of direction and frequency, could be an important characteristic of a bat's biosonar since they determine which part of an environment is illuminated with a given frequency and at a given time. Numerical results have predicted intricate geometrical features such as multiple peaks and notches in bat biosonar beampatterns. Furthermore, it has been shown that bats with nasal emission change the shapes of their emission baffles, “noseleaves,” during pulse emission. The work presented here is aimed at understanding if and how these two aspects (beam geometry and emission dynamics) are integrated into the bats' biosonar behaviors on the wing. The experiments are conducted with Old World leaf-nosed bats (Hipposideros pratti and Hipposideros armiger) that have a pronounced noseleaf dynamics. The animals fly in a tunnel that is instrumented with synchronized arrays of high-speed video cameras and microphones. The setup allows for precise track...

Content-Based Image Retrieval Features: A Survey

Content-Based Image Retrieval (CBIR) systems have been used for the searching of relevant images in various research areas. In CBIR systems features such as shape, texture and color are used. The extraction of features is the main step on which the retrieval results depend. Color features in CBIR are used as in the color histogram, color moments, conventional color correlogram and color histogram. Color space selection is used to represent the information of color of the pixels of the query image. The shape is the basic characteristic of segmented regions of an image. Different methods are introduced for better retrieval using different shape representation techniques; earlier the global shape representations were used but with time moved towards local shape representations. The local shape is more related to the expressing of result instead of the method. Local shape features may be derived from the texture properties and the color derivatives. Texture features have been used for images of documents, segmentation-based recognition,and satellite images. Texture features are used in different CBIR systems along with color, shape, geometrical structure and sift features.

Regression‐Based Landmark Detection on Dynamic Human Models

AbstractDetecting anatomical landmarks on various human models with dynamic poses remains an important and challenging problem in computer graphics research. We present a novel framework that consists of two‐level regressors for finding correlations between human shapes and landmark positions in both body part and holistic scales. To this end, we first develop pose invariant coordinates of landmarks that represent both local and global shape features by using the pose invariant local shape descriptors and their spatial relationships. Our body part‐level regression deals with the shape features from only those body parts that correspond to a certain landmark. In order to do this, we develop a method that identifies such body parts per landmark, by using geometric shape dictionary obtained through the bag of features method. Our method is nearly automatic, as it requires human assistance only once to differentiate the left and right sides. The method also shows the prediction accuracy comparable to or better than those of existing methods, with a test data set containing a large variation of human shapes and poses.

Different Measures of Structural Similarity Tap Different Aspects of Visual Object Processing.

The structural similarity of objects has been an important variable in explaining why some objects are easier to categorize at a superordinate level than to individuate, and also why some patients with brain injury have more difficulties in recognizing natural (structurally similar) objects than artifacts (structurally distinct objects). In spite of its merits as an explanatory variable, structural similarity is not a unitary construct, and it has been operationalized in different ways. Furthermore, even though measures of structural similarity have been successful in explaining task and category-effects, this has been based more on implication than on direct empirical demonstrations. Here, the direct influence of two different measures of structural similarity, contour overlap and within-item structural diversity, on object individuation (object decision) and superordinate categorization performance is examined. Both measures can account for performance differences across objects, but in different conditions. It is argued that this reflects differences between the measures in whether they tap: (i) global or local shape characteristics, and (ii) between- or within-category structural similarity.

Shape Based Image Retrieval using Lower Order Zernike Moments

Shape is one of the significant features of Content Based Image Retrieval (CBIR). This paper proposes a strong and successful shape feature, which is based on a set of orthogonal complex moments of images known as Zernike moments. For shape classification Zernike moment (ZM) is the dominant solution. The radial polynomial of Zernike moment produces the number of concentric circles based on the order. As the order increases number of circles will increases, due to this the local information of an image will be ignored. In this paper, we introduced a novel method for radial polynomial where local information of an image given importance. We succeeded to extract the local features and shape features at very a low order of polynomial compared to the state of traditional ZM.The proposed method gives an advantage of a lower order, less complex, and lower dimension feature vector.For more similar images we find that simple Euclidian distance approximately zero. Proposed method tested on a MPEG-7 CE-1 shape database, Coil-100 databases. Experiments demonstrated that it is outperforming in identifying the shape of an object in the image and reduced the retrieving time and complexity of calculations.

Multi-attribute statistics histograms for accurate and robust pairwise registration of range images

Registration of range images based on local shape features is widely adopted due to its validated effectiveness and robustness, while most existing local shape descriptors struggle to simultaneously achieve a pleasurable and balanced performance in terms of distinctiveness, robustness and time efficiency. This paper proposes a novel representation of 3D local surfaces, called multi-attribute statistics histograms (MaSH), for automatic registration of range images. MaSH comprises both spatial and geometric information characterizations. The characterization of spatial information is achieved via radial partitions in the 3D local support volume around the keypoint based on a local reference axis (LRA), creating a set of subspaces. While the encoding the shape geometry is performed by calculating statistical histograms of multiple faint correlated geometric attributes (i.e., local depth, normal deviation, and surface variation angle) for each subspace, so as to achieve information complementarity. Then, a robust rigid transformation estimation algorithm named 2-point based sample consensus with global constrain (2SAC-GC) is presented to tackle the problem of calculating an optimal transformation from the correspondence set with outliers. Finally, a coarse-to-fine registration method based on MaSH and 2SAC-GC is proposed for aligning range images. Experiments on both high-resolution (Laser Scanner) and low-resolution (Kinect) datasets report that, our method achieves a registration accuracy of 90.36% and 80.39% on the two datasets, respectively. It also exhibits strong robustness against noise and varying mesh resolutions. Furthermore, feature matching experiments show the over-all superiority of the proposed MaSH descriptor against the state-of-the-arts including the spin image, snapshots, THRIFT, FPFH, RoPS, LFSH and RCS descriptors.

An Effective 3D Shape Descriptor for Object Recognition with RGB-D Sensors

RGB-D sensors have been widely used in various areas of computer vision and graphics. A good descriptor will effectively improve the performance of operation. This article further analyzes the recognition performance of shape features extracted from multi-modality source data using RGB-D sensors. A hybrid shape descriptor is proposed as a representation of objects for recognition. We first extracted five 2D shape features from contour-based images and five 3D shape features over point cloud data to capture the global and local shape characteristics of an object. The recognition performance was tested for category recognition and instance recognition. Experimental results show that the proposed shape descriptor outperforms several common global-to-global shape descriptors and is comparable to some partial-to-global shape descriptors that achieved the best accuracies in category and instance recognition. Contribution of partial features and computational complexity were also analyzed. The results indicate that the proposed shape features are strong cues for object recognition and can be combined with other features to boost accuracy.

身体部位の特徴点と形状情報に基づくモデルベース歩容認証の検討

Gait Recognition is recently evolving techniques by which we can recognize individuals by one's gait. There are two major approaches; silhouette-based and model-based. In Japan, a method based on GEI (Gait Energy Image), which is one of the silhouette-based approaches, is beginning to be used for forensic purposes. It is a problem, however, that person's silhouettes' variabilities due to conditions such as view angles, resolutions, frame-rates, and body regions used for analysis, and so on, sometimes lessen recognition reliability under the use of GEI method only. It is an urgent task to expand the ranges of application. In order to try to resolve the task, it could be valid to use several different approaches, including model-based ones. Previous studies of model-based methods can be divided in two types; manual and automated. However, it is a problem that manual methods require a person of special skills and experiences, and automated methods only can apply specific view condition, so it is required to develop finer model-based methods. Here, we proposed a novel model-based-like method based on feature points (resemble body joints) and local shape features around feature points. We examined error rates of the proposed method under various conditions of frame-rates, resolutions, and regions used in analysis, and compared them with those of GEI method. We found that the proposed method showed a little higher recognition rate than GEI one when captured from lateral direction and when upper body is used in analysis, and showed less dependency on low frame-rate conditions than GEI one when captured from lateral direction.

Action recognition with low observational latency via part movement model

In this paper, we address the issue of recognizing human actions with low observational latency, which is vital for many applications such as virtual reality and interactive entertainment. Then our purpose is to achieve competitive action recognition performance from very short video clips. Such a task essentially is challenging because only very limited information is provided. To this end, we first develop a feature extraction method to exploit both motion (local flow) and appearance (local shape) features such that the information insufficiency can be effectively mitigated. Then we propose an action representation method named Part Movement Model (PMM), which explicitly captures the spatial-temporal structure of human actions and divides the actions into discriminative part movements. Consequently, the actions can be better represented and the competitive performance can be achieved although only short clips are used. Finally, we experimentally verify the effectiveness of the proposed method on three benchmark datasets. The results show that short clips of 6−7 frames (0.2−0.3 second video) are enough to achieve the recognition performance comparable to the baselines with high latency.

Compressive Buckling of Rippled Graphene via Molecular Dynamics Simulations

This paper studies the compressive buckling of near-square rippled monolayer graphene sheets in thermal environments by using molecular dynamics simulations. Armchair and zigzag graphene sheets are both considered and the four edges of a graphene sheet are either simply supported or clamped. Compressive force instead of the commonly used compressive strain is gradually applied to the edges of the graphene sheet until the graphene sheet is crushed. It is found that ripples in the graphene are formed owing to thermal fluctuations even before the compressive force is applied. The amplitude of ripples is reduced when the compressive force is applied to the graphene. The buckling mode of the graphene may show local or global mode shape features, depending on the chirality, the support conditions and the size of the graphene. Most buckling modes of the graphene sheets are very different from the global (1,1) mode of a uniaxially loaded and simply supported or clamped square plate as predicted by continuum mechanics models. A clamped graphene can take significantly more compressive force after the occurrence of initial buckling and before the graphene is completely crushed. A zigzag graphene has a larger initial buckling force than its armchair counterpart. Raising the environmental temperature may either increase or decrease the initial buckling force of the graphene depending on the support conditions and the chirality of the graphene.

Integration of emitter and receiver dynamics in a biomimetic sonar head

Horseshoe bats (family Rhinolophidae) transmit and receive sonar information about the environment using morphologically complex and dynamic noseleaf and pinnae. Both local shape features and dynamic deformations of the noseleaf and pinnae contribute to properties of the bat biosonar system. This work systematically explores the effect of local shape features on the dynamic properties of emission and reception baffles using a biomimetic robotic system inspired by the horseshoe bat. Parametric models of the greater horseshoe bat (Rhinolophus ferrumequinum) noseleaf and pinnae with selectable combinations of local shape features were developed. Baffles with feature realizations were cast out of flexible silicone and mounted to a platform to dynamically actuate the emission and reception baffle surfaces with motions patterns similar to the greater horseshoe bat. Motions of the baffle surfaces were synchronized to the outgoing and incoming sonar waveform, and the time-frequency properties of the emission and ...

Symmetry aware embedding for shape correspondence

In this paper, we present symmetry-aware embedding for shape correspondence, which is robust against symmetric (left–right) flips and rotational (front–back) flips. Unlike previous embedding approaches that embed surfaces into a high-dimensional space, our technique is based on a low dimensional (3D) embedding. Our method can solve left–right flips by finding a 3D rigid transformation between two embedding surfaces without reflections. Using the global reflectional symmetry plane to align two surfaces, we can further reduce the problem to that of finding the rotation that corrects the signs of the front–back and up–down directions (there are four possible solutions). We exploit this simple problem formulation and alleviate the front–back flips, by explicitly comparing the front and back of embedding surfaces based on the global and local extrinsic shape characteristics. Consequently, reasonably accurate point-to-point correspondences can be established simply by performing the nearest neighbor search in our embedding space. Experimental results based on a shape correspondence benchmark showed that our method produces stable matching results.

Spatial Feature Extractions to Reduce Intra-Class Variability in Traffic Sign Templates

This paper presents shape features and a normalization scheme to address intra-class variation challenge of various traffic sign template designs. Traffic sign design around the world must conform to certain guidelines. However, there may be spatial differences among traffic signs belonging to a same class. In this work, we propose 22 features to discriminate 23 classes of traffic signs. This low number of features would minimize the classification time. We highlight the usage of location and directional information of local shape features. The application of the features is explained using the Naive Bayes classifier for classification of the images. A dataset containing traffic sign templates with up to a maximum of 22 countries per class was used in our experiment to simulate intra-class variations. The features were compared against twelve other classifiers. The Naive Bayes classification technique performs the best with 99.4% classification accuracy and average of 0.43ms classification time per feature set. We proved that the features proposed are effective to discriminate inter-class images even though there are intra-class differences.