Contour Detector Research Articles

Learning to refine object boundaries

Existing Deep-Learning-based contour detectors suffer from the issues of the sharpness and the correctness of their predictions, which often need offline post-processing to sharpen the results as well as improve model performance. In this work, we present a novel method that can learn to refine object contour in training and directly output crisp object boundaries in inference. To this end, we first introduce a keypoint-focal loss that draws point-based attention to the isolated contour annotations. It allows an edge detector to jointly optimize the appearance thickness and the localization accuracy of predictions in the training procedure. Moreover, we present a regularization loss to further improve the performance of an edge detector. Lastly, we present the Contour Transformer model for precisely localizing object boundaries in images. We repurpose and integrate a Transformer-style hyper module into an encoder–decoder network, effectively aggregating global contextual information on high-level features and significantly enhancing the discriminative power for classifying foreground/background pixels. We train and test our Attention and Contour Transformer detector (ACTD) on four widely adopted datasets, i.e., BSDS500, NYUD, Multi-Cue, and RoadNet. The proposed method achieves an ODS F-score of 0.826 on BSDS500 and an ODS F-score of 0.783 on NYUD, outperforming previous top detectors.

A hybrid approach for face recognition using a convolutional neural network combined with feature extraction techniques

Facial recognition technology has been used in many fields such as security, biometric identification, robotics, video surveillance, health, and commerce due to its ease of implementation and minimal data processing time. However, this technology is influenced by the presence of variations such as pose, lighting, or occlusion. In this paper, we propose a new approach to improve the accuracy rate of face recognition in the presence of variation or occlusion, by combining feature extraction with a histogram of oriented gradient (HOG), scale invariant feature transform (SIFT), Gabor, and the Canny contour detector techniques, as well as a convolutional neural network (CNN) architecture, tested with several combinations of the activation function used (Softmax and Segmoïd) and the optimization algorithm used during training (adam, Adamax, RMSprop, and stochastic gradient descent (SGD)). For this, a preprocessing was performed on two databases of our database of faces (ORL) and Sheffield faces used, then we perform a feature extraction operation with the mentioned techniques and then pass them to our used CNN architecture. The results of our simulations show a high performance of the SIFT+CNN combination, in the case of the presence of variations with an accuracy rate up to 100%.

3D map building based on extrinsic sensor calibration method and object contour detector with a fully convolutional neural network

In this paper, we propose a simple method to obtain an object’s 3D coordinate information in one image, using one monocular camera and two 2D LiDAR sensors, which are widely used low-cost sensors. An extrinsic calibration method is used for each LiDAR sensor to transfer LiDAR coordinate to camera pixel coordinate in one image. The proportion factor [Formula: see text] is calculated, based on the relationship of pixel from calibration points and distance from detected points of two LiDAR sensors. In order to create a correct 3D map, a deep learning algorithm for real-time object contour detection with a fully Convolutional Neural Network is proposed, through gathering data, labeling data, training model, and testing object contour detector. For 3D map building, by the proportion factor [Formula: see text] and the detector contour feature pixels, the height of the object is obtained. Combining with LiDAR sensor’s detection information, the 3D posture of the object is obtained. Based on the proposed extrinsic sensor calibration method and the object contour detect result, virtual points of the object are calculated. Finally, the experiment results show the 3D coordinate information of the detected single or multiple objects. Assembling sensors and a controller on a Kobuki robot, the indoor 3D map is built based on the proposed methods.

Bidirectional Multiscale Refinement Network for Crisp Edge Detection

With the development of deep convolution neural networks (CNNs), contour detection has made great progress. Some contour detectors based on CNNs have better performance than human beings on standard benchmarks. However, it is easier for CNNs to learn the similar features of adjacent pixels, and the number of background pixels and edge pixels in the input training sample is highly imbalanced. Therefore, the prediction edge by the edge detector based on CNNs is thick and requires post-processing to obtain crisp edges. Accordingly, we introduce a novel parallel attention model and a novel loss function that combines cross-entropy and dice loss through the use of adaptive coefficients, and propose a novel bidirectional multiscale refinement network (BMRN) that stacks multiple refinement modules in order to achieve richer feature representation. The experimental results show that our method has better performance than the state-of-the-art on BSDS500 (ODS F-score of 0.828), NYUDv2 depth datasets (ODS F-score of 0.778) and Multi-Cue dataset (ODS F-score 0.905(0.002)).

Drone-Assisted Image Processing Scheme using Frame-Based Location Identification for Crack and Energy Loss Detection in Building Envelopes

This paper presents improved methods to detect cracks and thermal leakage in building envelopes using unmanned aerial vehicles (UAV) (i.e., drones) with video camcorders and/or infrared cameras. Three widely used contour detectors of Sobel, Laplacian, and Canny algorithms were compared to find a better solution with low computational overhead. Furthermore, a scheme using frame-based location identification was developed to effectively utilize the existing approach by finding the current location of the drone-assisted image frame. The results showed a simplified drone-assisted scheme along with automation, higher accuracy, and better speed while using lower battery energy. Furthermore, this paper found that the cost-effective drone with the attached equipment generated accurate results without using an expensive drone. The new scheme of this paper will contribute to automated anomaly detection, energy auditing, and commissioning for sustainably built environments.

Autonomous Vision-Guided Object Collection From Water Surfaces With a Customized Multirotor

This article presents a multirotor system with an integrated net mechanism for autonomous object detection and collection from water surfaces utilizing only onboard sensors. The task of object collection on water surfaces is challenging due to the following reasons: 1) propeller outwash alters the object's dynamics; 2) unpredictable current flows, 3) extreme reflection and glare off water surfaces affect the object detection; and 4) noisy height measurements over water surface. A two-phase object-detection algorithm is developed with a linearized polarization filter and specularity removal algorithm to eliminate reflections and an edge-based contour detector for detection of objects on water surface. Subsequently, a boundary layer sliding-mode control is implemented to ensure the system is robust to modeling uncertainties. A dynamic sliding surface is designed based on constrained linear model predictive control. The efficacy of the proposed collection system is validated by multiple outdoor tests. Multiple objects of different shapes and sizes are collected with an overall success rate of 91.6%.

Lateral refinement network for contour detection

There has been significant progress in contour detection with the development of convolutional neural networks. To improve the detection performance of the object contour in the complex scenes, we proposed a novel lateral refinement network (LRNet) that extracts feature information from multiple refinement modules. In addition, we modified LRNet to train an effective contour detector, called lateral refinement contour (LRC). LRNet expands the generic refinement architecture and has different refinement levels that make the refinement network deeper to extract richer convolutional features. The low-level refinement modules explicitly exploit information available along the side output of the down-sampling process, and the high-level refinement modules attempt to fuse the low-level output features. The proposed method improves the performance of image-to-image predictions by deep-stacking all the meaningful refinement levels in a holistic manner. Using the pre-trained VGG16 network and Resnet network as backbones to train LRNet, we achieved state-of-the-art performance on several available datasets. The experimental results on the BSDS dataset are ODS = 0.816 and ODS = 0.820, and the experimental results on the NYUD-v2 dataset are ODS = 0.761 and ODS = 0.760. Especially, the results on BSDS500 surpassed the human-level performance under stricter criteria.

A robust contour detection operator with combined push-pull inhibition and surround suppression

Contour detection is a salient operation in many computer vision applications as it extracts features that are important for distinguishing objects in scenes. It is believed to be a primary role of simple cells in visual cortex of the mammalian brain. Many of such cells receive push-pull inhibition or surround suppression. We propose a computational model that exhibits a combination of these two phenomena. It is based on two existing models, which have been proven to be very effective for contour detection. In particular, we introduce a brain-inspired contour operator that combines push-pull and surround inhibition. It turns out that this combination results in a more effective contour detector, which suppresses texture while keeping the strongest responses to lines and edges, when compared to existing models. The proposed model consists of a Combination of Receptive Field (or CORF) model with push-pull inhibition, extended with surround suppression. We demonstrate the effectiveness of the proposed approach on the RuG and Berkeley benchmark data sets of 40 and 500 images, respectively. The proposed push-pull CORF operator with surround suppression outperforms the one without suppression with high statistical significance.

Delineation of Agricultural Field Boundaries from Sentinel-2 Images Using a Novel Super-Resolution Contour Detector Based on Fully Convolutional Networks

Boundaries of agricultural fields are important features necessary for defining the location, shape, and spatial extent of agricultural units. They are commonly used to summarize production statistics at the field level. In this study, we investigate the delineation of agricultural field boundaries (AFB) from Sentinel-2 satellite images acquired over the Flevoland province, the Netherlands, using a deep learning technique based on fully convolutional networks (FCNs). We designed a multiple dilation fully convolutional network (MD-FCN) for AFB detection from Sentinel-2 images at 10 m resolution. Furthermore, we developed a novel super-resolution semantic contour detection network (named SRC-Net) using a transposed convolutional layer in the FCN architecture to enhance the spatial resolution of the AFB output from 10 m to 5 m resolution. The SRC-Net also improves the AFB maps at 5 m resolution by exploiting the spatial-contextual information in the label space. The results of the proposed SRC-Net outperform alternative upsampling techniques and are only slightly inferior to the results of the MD-FCN for AFB detection from RapidEye images acquired at 5 m resolution.

Interest point detection method based on multi‐scale Gabor filters

In this study, a novel interest point detection algorithm which combines image intensity variation and edge contour information is proposed. Firstly, the Canny edge contour detector is used to extract the edge map. Secondly, the imaginary parts of multi-scale Gabor filters are applied to smooth input image and then the normalised information entropies at various scales can be acquired. Finally, multiplication of different normalised information entropies will be served as a new measure for interest point that can also be used for interest point detection. This method has two advantages: on the one hand, detection accuracy is greatly improved because combination information is adopted to extract interest points including contour shape information, grey variation of edge pixels and their neighbours; on the other hand, non-interest points can be well inhabited due to multi-scale product in detector is served as an interest point measure. Also, desirable noise robustness and time efficiency are validated through experiments. Compared with four other state-of-art methods, the proposed method shows excellent performance in terms of geometric transformations and localisation accuracy of repeated interest points.

Semi-supervised map regionalization for categorical data

ABSTRACTThe objective of map regionalization is to group contiguous objects on a map into larger entities sharing similar properties or relationships, resulting in homogeneous regions that are easier to interpret. We propose a strategy to interactively incorporate human perception of homogeneous regions to improve unsupervised regionalization processes. The approach fits within the well-known segmentation/clustering framework. The method operates on a categorical map, introduces a contour detector for boundaries delineation with better resolution power than a regular grid tessellation to initiate a region growing process, and integrates the role of a human analyst for better classification of homogeneous areas through a semi-supervised clustering (SSC) method. This last step is achieved using pairwise clustering constraints on regions identified by the analyst on the monitor. The potential of the proposed strategy is illustrated with data extracted from the Earth Observation for the sustainable development of forests (EOSD) map of Canada. Comparisons with a recently introduced algorithm for map regionalization are provided for three different spatial scales at different steps of the method.

Pedestrian detection using multiple feature channels and contour cues with census transform histogram and random forest classifier

This paper presents a reliable and real-time method to detect pedestrians in image scenes that can vary greatly in appearance. To achieve greater reliability in what can be detected, a combination of visual cues is used in conjunction with edge-based features and colour information as a basis for training a random forest (RF) classifier to detect the local contour cues for pedestrian images. To achieve a real-time detection rate, the contour cues, edge-based features and colour information are incorporated and then trained using a cascade RF classifier with a census transform histogram visual descriptor that implicitly captures the global contours of the pedestrians. The contour detector favourably exceeded previous leading contour detectors and achieved a 95% detection rate. The reliability and specificity of the pedestrian detector are demonstrated on more than 5000 positive images containing street furniture, lamp posts and trees, structures that are frequently confused with persons by computer vision systems. Evaluation with over 220 video sequences with $$640 \times 480$$ pixel resolution presented a true positive rate of 96%. The proposed pedestrian detector outperforms previous competitive pedestrian detectors on many varied person data sets. The speed of execution in a robot is about 62 ms per frame for images of $$640 \times 480$$ pixels on an Intel Core i3-2310M™ processor running at 2.10 GHz with a RAM of 4 GB.

High throughput automatic muscle image segmentation using parallel framework

BackgroundFast and accurate automatic segmentation of skeletal muscle cell image is crucial for the diagnosis of muscle related diseases, which extremely reduces the labor-intensive manual annotation. Recently, several methods have been presented for automatic muscle cell segmentation. However, most methods exhibit high model complexity and time cost, and they are not adaptive to large-scale images such as whole-slide scanned specimens.MethodsIn this paper, we propose a novel distributed computing approach, which adopts both data and model parallel, for fast muscle cell segmentation. With a master-worker parallelism manner, the image data in the master is distributed onto multiple workers based on the Spark cloud computing platform. On each worker node, we first detect cell contours using a structured random forest (SRF) contour detector with fast parallel prediction and generate region candidates using a superpixel technique. Next, we propose a novel hierarchical tree based region selection algorithm for cell segmentation based on the conditional random field (CRF) algorithm. We divide the region selection algorithm into multiple sub-problems, which can be further parallelized using multi-core programming.ResultsWe test the performance of the proposed method on a large-scale haematoxylin and eosin (H &E) stained skeletal muscle image dataset. Compared with the standalone implementation, the proposed method achieves more than 10 times speed improvement on very large-scale muscle images containing hundreds to thousands of cells. Meanwhile, our proposed method produces high-quality segmentation results compared with several state-of-the-art methods.ConclusionsThis paper presents a parallel muscle image segmentation method with both data and model parallelism on multiple machines. The parallel strategy exhibits high compatibility to our muscle segmentation framework. The proposed method achieves high-throughput effective cell segmentation on large-scale muscle images.

ОЦЕНКА ЭФФЕКТИВНОСТИ КОНТУРНЫХ ДЕТЕКТОРОВ НА ОСНОВЕ ТОЧЕЧНОГО ПОТОКА ВОССТАНОВЛЕНИЯ

Актуальность. В настоящее время много внимания уделяется созданию эффективных систем, предназначенных для обработки, хранения и передачи больших потоков информации, получаемой в виде изображений пространственно-временных полей в таких областях человеческой деятельности, как космонавтика, геология и геофизика, картография, навигация, дефектоскопия и во многих других. В большинстве этих систем при реализации алгоритмов детектирования в качестве эталонных изображений используются реальные снимки. Например, это характерно для выделения и прослеживания границ отраженных волн на изображениях сейсмических полей, получаемых при поиске нефтяных и газовых месторождений. Представление эталонных изображений в виде реальных снимков позволяет настроить алгоритм на конкретную предметную область, но осложняет оценку качества используемого алгоритма и проведение его объективного сравнения с другими алгоритмами, реализующими подобные процедуры. Цель: исследование эффективности выделения контурного рисунка изображений различными алгоритмами детектирования на основе обобщённого комплексного критерия, предложенного авторами данной работы. Это позволит подойти объективно к выбору наилучшего алгоритма детектирования и подбору оптимальных значений его параметров для любой предметной области. Методы и средства исследования. При проведении экспериментов были использованы методы стохастического моделирования пространственно-временных сигналов и полей. Сам вычислительный эксперимент проводился в среде программного комплекса «КИМ СП». Результаты. Проведен анализ методов оценивания операторов поиска и локализации границ, и обобщены основные типы ошибок. На основе предложенной классификации метрик эффективности работы контурных детекторов рассмотрен обобщённый комплексный критерий оценки качества выделения контурных рисунков в изображениях, показаны возможные составляющие обобщённого показателя качества и формальные подходы к выбору весовых коэффициентов метрик. Приведён пример реализации обобщенного показателя качества, включающего пять метрик, и исследована его эффективность для трех квазиоптимальных алгоритмов оконтуривания («Canny», «Marr» и «ISEF») при различных уровнях шума. Проведенные исследования позволили сделать выводы об объективности полученных результатов и дать рекомендации по применению алгоритмов оконтуривания.

Automatic Human Sperm Concentrartion in microscopic videos


 Background: Human sperm cell counting analysis is of significant interest to biologists studying sperm function and to medical practitioners evaluating male infertility. Currently the analysis of this assessment is done manually by looking at the sperm samples through a phase-contrast microscope using expert knowledge to do a subjective judgement of the quality.
 Aims: to eliminate the subjective and error prone of the manual semen analysis and to avoid inter and intra-laboratory inconsistencies in semen analysis test results
 Methods: In this paper we introduce a technique for human sperm concentration. Its principle is based on the execution of three steps: The first step in unavoidable. It concerns the pretreatment of the human sperm microscopic videos which consists of a conversion of the RGB color space into the YCbCr space, the “Gaussian filtering” and the “discrete wavelet filtering”. The second step is devoted to the segmentation of the image into two classes: spermatozoas and the background. To achieve this, we used an edge detection technique “Sobel Contour detector”. The third step is to separate true sperm from false ones. It uses a machine learning technique of type decision trees that consist on two classes classification based on invariant characteristics that are the dimensions of the bounding ellipse of the spermatozoid head as well as its surface.
 Results: To test the robustness of our system, we compared our results with those performed manually by andrologists. After results analysis, we can conclude that our system brings a real improvement of precision as well as treatment time which make it might be useful for groups who intend to design new CASA systems.
 Conclusion: In this study, we designed and implemented a system for automatic concentration assessment based on machine learning method and image processing techniques.

FEATURE VECTOR GENERATION FOR THE FACIAL EXPRESSION RECOGNITION USING NEO-FUZZY SYSTEM

Context. The article is devoted to the problem of a training data set forming for the automatic human emotions recognition system on the basis of a multidimensional extended neo-fuzzy neuron. The aspects of choice the attributes vector’s dimension and composition, their influence on the system learning rate are considered. The object of research is the method of multidimensional data clustering. The subject of research is two-dimensional images geometric features systematization. Objective. The main goal of the work is to develop an approach to person’s face expression description using geometric features fixed set that can be obtained by video sequence frames processing. Method. To study the facial expressions recognition system it is proposed to form a feature vector consisting of characteristic points coordinates. There were selected points that relate to the location and shape of the eyelids, eyebrows, eye pupils, lips contours, nose wings, nasolabial folds. Such points can be easily found during the automatic image processing using known contour detectors. Also, the possibility of using for the human facial expression description not the coordinates of characteristic points, but the distances between them, was investigated. From these distances a different feature vector was created, the properties of which were compared with the points coordinates vector. Results. The developed recognition system on the basis of a multidimensional extended neo-fuzzy neuron have been implemented in software and investigated for solving the problem of facial expression classification. A comparison between the attribute vectors that are different in composition and dimension is made. The structure for the feature vector, which provides high system learning rate, and does not require the additional structural elements was chosen. Conclusions. The experimental study fully confirms the effectiveness of the developed approach for the human facial expressions recognition using a multidimensional extended neo-fuzzy neuron.

Contour detection model using linear and non‐linear modulation based on non‐CRF suppression

Psychophysical and neurophysiological investigations on the human visual system show that most neurons in the primary visual cortex (V1) possess a non-classical receptive field (nCRF) region in addition to the CRF region. The nCRF has a modulatory, normally inhibitory, effect on the responses to visual stimuli generated within the CRF. In computational terms, this mechanism suppresses the response to edges in the presence of similar edges in the surroundings. Many computational techniques have been proposed to address the surround suppression mechanism. These methods introduce an inhibition term that is required to suppress the textures and protect the contours. Several studies have found that the spatial summation properties over the receptive fields of retinal X cells are approximately linear, while they are non-linear for Y cells. Inspired by the visual information processing in the X–Y channel and spatial summation properties of X and Y cells, the authors propose a contour detector using linear and non-linear modulations based on nCRF suppression. Extensive experimental evaluations demonstrate that their contour detector significantly outperforms other algorithms. The methods proposed in this study are expected to facilitate the development of efficient computational models in the field of machine vision.

Role of zero synapses in unsupervised feature learning

Synapses in real neural circuits can take discrete values including zero (silent or potential) synapses. The computational role of zero synapses in unsupervised feature learning of unlabeled noisy data is still unclear, thus it is important to understand how the sparseness of synaptic activity is shaped during learning and its relationship with receptive field formation. Here, we formulate this kind of sparse feature learning by a statistical mechanics approach. We find that learning decreases the fraction of zero synapses, and when the fraction decreases rapidly around a critical data size, an intrinsically structured receptive field starts to develop. Further increasing the data size refines the receptive field, while a very small fraction of zero synapses remain to act as contour detectors. This phenomenon is discovered not only in learning a handwritten digits dataset, but also in learning retinal neural activity measured in a natural-movie-stimuli experiment.

Evaluation of Hierarchical Watersheds.

This paper aims to understand the practical features of hierarchies of morphological segmentations, namely the quasi-flat zones hierarchy and watershed hierarchies, and to evaluate their potential in the context of natural image analysis. We propose a novel evaluation framework for the hierarchies of partitions designed to capture various aspects of those representations: precision of their regions and contours, possibility to extract high quality horizontal cuts and optimal non-horizontal cuts for image segmentation, and the ease of finding a set of regions representing a semantic object. This framework is used to assess and to optimize hierarchies with respect to the possible pre- and post-processing steps. We show that, used in conjunction with a state-of-the-art contour detector, watershed hierarchies are competitive with the complex state-of-the-art methods for hierarchy construction. In particular, the proposed framework allows us to identify a watershed hierarchy based on a novel extinction value, the number of parent nodes that outperforms the other hierarchies of morphological segmentations. This coupled with the fact that watershed hierarchies satisfy clear global optimality properties and can be efficiently computed on large data, make them valuable candidates for various computer vision tasks.

Comparative Study of Harris and Active Contour using Viola-Jones Algorithm for Facial Landmarks Detection

In this paper, we present a comparative study of two methods: Harris corners detector (H) and Active Contour detector (A.C) using Viola-Jones algorithm (V.J) for facial landmarks (eyes, nose, mouth) corners detection. These methods were implemented on two face databases; ECVP and FEI databases with combination of two methods (V.J + H) and (V.J + A.C). Experimental results showed that (V.J +A.C) gives a higher rate of detection than (V.J + H) method.