Search

Spondyloarthritis (SpA) is a group of inflammatory diseases that cause severe damage in the structure of the skeleton. One of the most important features that are assessed in the diagnosis of SpA disorders and during the monitoring of the progression of this disease is the shape of the vertebrae and the appearance of bony outgrowths in the spine region. For this purpose, radiography is often used. This paper presents a novel automated method of external contour segmentation of vertebrae in X-ray images. The proposed algorithm consists of preprocessing, edge detection using the Sobel operator, binarization, area opening, and skeletonization. During the study, the impact of different filters (i.e., Gaussian filter, sigma filter, and anisotropic diffusion filter) on the quality of the results was also investigated. The method’s efficiency was tested on a dataset containing 11 lateral cervical spine radiographs. The results show that the method has the potential to become an automatic tool used by physicians to determine the shape of vertebrae.

Vessel centerline extraction is fundamental for plentiful medical applications. Majority of current methods require pre-segmentations, distance maps or similar sorts of scanning whole volume action and followed by minimal-path or skeletonization algorithms. In this paper, we demonstrate a deep reinforced tree-traversal agent that automatically traces tree-structure centerlines assuming no post-prune or post-merging. It takes raw images as input and generates tree-structure centerlines naturally. To this end, road mark and dynamic reward mechanisms are proposed to make tree-structure vessels learnable and impart the agent how to learn correspondingly. Besides, a multi-task discriminator is raised to simultaneously detect bifurcations and decide terminations. We experimentally show that traced centerlines have an overlap of more than 90% and a distance less than 0.25 mm with annotated reference centerlines on coronary arteries. Beyond the promising accuracy, the proposed method also surpasses other existing methods by a large margin in terms of the time and memory efficiency. And a flexible trade-off between accuracy and time efficiency is exhibited at the inference. Codes are available at https://github.com/LzVv123456/Deep-Reinforced-Tree-Traversal.

Strand proposed a distance-based sequential thinning algorithm for producing surface skeletons of binary objects sampled on the face-centered cubic (FCC) grid. In this paper, we present two modified versions of his algorithm, which are faster and - according to our experiments - less sensitive to the visiting order of border points in the sequential thinning phase.

Nowadays one of the most popular topics in fingerprint recognition academic research is the high-resolution fingerprint identification way. This technique has been prominently attractive to the worldwide scientific community thanks to the possibility of using level 3 features like pores which cannot be detected in lower resolution images. In this context, this chapter proposes two contributions: First, a pore detection method for high-resolution fingerprint image based on the morphological operation (skeletonization) and the labeled connect component method. Second, a new method to match pores without an alignment. Our matching approach is based on the contextual characteristics of the pore. It consists of positions and orientations of pore neighbors, which are defined as polar coordinates given in the polar system centered on the considered pore. The proposed algorithms are tested on a high-resolution fingerprint database. Experimental results show that our methods outperform the existing algorithms.

The incisions and the subsequent skeletonization of the cartilage and bone nasal structures, in the great majority of rhinoplasty cases, must be carried out symmetrically and with extreme precision. Achievement of the correct dissection planes must be immediate and maintained throughout the operation. Any wrong gesture and any error in flap elevation can cause complications such as nasal stenosis, visible irregularities, and residual asymmetries.

International audience; Nowadays, the security of both digital and hardcopy documents has become a real issue. As a solution, numerous integrity check approaches have been designed. The challenge lies in finding features which are robust to print-and-scan process. In this paper, we propose a new method of printed-and-scanned character matching based on the adaptation of biometrical features. After the binarization and the skeletonization of a character, feature points are extracted by computing crossing numbers. The feature point set can then be smoothed to make it more suitable for template matching. From various experimental results, we have shown that an accuracy of more than 95% is achieved for printand-scan resolutions of 300 dpi and 600 dpi. We have also highlighted the feasibility of the proposed method in case of double print-and-scan operation. The comparison with a state-of-the-art method shows that the generalization of proposed matching method is possible while using different fonts.

This document explores an autonomous navigation approach based on the generation of free-obstacle navigation diagrams by means of using the skeletonization technique and other image processing algorithms. These diagrams are turned into a path by means of a vector of points and performed by a hybrid robotic platform, composed of a humanoid robot and one with omni-directional locomotion. Four different configurations (origin, obstacles and destination) were studied in order to test the algorithm performance, computing the mean squared error and the average of relative error for each point in the path. For all of the tested configurations, the proposed algorithm chooses the shortest path through the navigation diagram. The final point and mean point-by-point relative errors were obtained for each of the 4 proposed configurations, where the average values over all of them were 1.11% for the final point relative error and 1.23% for the mean relative error, which validates the use of the algorithm for the autonomous generation of navigation paths for the hybrid robotic platform under study.

Thinning is a frequently used approach to produce all kinds of skeleton-like shape features in a topology-preserving way. It is an iterative object reduction: some border points of binary objects are deleted, and the entire process is repeated until stability is reached. In the conventional implementation of thinning algorithms, we have to investigate the deletability of all border points in each iteration step. In this paper, we introduce the concept of k-attempt thinning (\(k\ge 1\)). In the case of a k-attempt algorithm, if a border point ‘survives’ at least k successive iterations, it is ‘immortal’ (i.e., it belongs to the produced feature). We propose a computationally efficient implementation scheme for k-attempt thinning. It is shown that an existing parallel thinning algorithm is 5-attempt, and the advantage of the new implementation scheme over the conventional one is also illustrated.

Character segmentation is the key to CAPTCHA recognition. In order to solve the problem of low success rate of CAPTCHA segmentation caused by adhesive characters, an adhesion character segmentation algorithm based on corner detection and K-Means clustering was proposed. The algorithm performs corner detection on the CAPTCHA image of the adhesive characters, then uses K-Means clustering method to cluster the corner points of ROI, and determines the adhesion character segmentation line from the clustering results. The experimental results are compared with the drop-fall and skeletonization, in which the recognition accuracy of the image with serious adhesion is 92%. The result shows the superiority of the segmentation algorithm and provides a new method for the segmentation of adhesive characters.

Handwritten character recognition (HCR) is one of the significant issues in today’s emerging world. It is very difficult to identify the characters from a handwritten document using optical character recognition (OCR) technique. In our work, geometrical feature extraction and neural network computational algorithm are used for recognizing the offline handwritten Sanskrit characters. Initially the binarization and denoising processes are performed on the scanned handwritten document. Later, skeletonization, skewness detection, and correction processes are performed. Image is segmented and required features are extracted and fed into the different classifiers for character recognition. Then, the comparative study of the Sanskrit character recognition is done by employing the RCS with BPNN, BPNN with RBF, and MLP. The proposed character recognition system deploys precision, mean square error rate, recall, false error rate, false-positive error rate, sensitivity, specificity, and accuracy for effective analysis of the handwritten Sanskrit characters.

Cerebral arteriovenous malformations (AVM) are prone to rupture, which will lead to life-threatening conditions. Because of the complexity and high mortality and disability rate of AVM, it has been a severe problem in surgery for many years. In this paper, we propose a new method of AVM location and segmentation based on graph theory. A weighted breadth-first search tree is created from the result of vascular skeletonization, and the AVM is automatically detected and extracted. The feeding arteries, draining veins and the AVM nidus are segmented according to the topological structure of the vessel. We evaluate the proposed method on clinical data sets and achieve an average accuracy of 95.14%, sensitivity of 82.28% and specificity of 94.88%. The results show that our method is effective and is helpful for the treatment of vascular interventional surgery.

In osteoporotic patients, the sites most prone to fracture are characterized by a predominance of cancellous bone. It has been shown that trabecular architecture plays a crucial role in the mechanical performance of this type of bone. It therefore appears necessary to be able to describe this architecture within an elementary volume. This requires a quantification of the anisotropy of the bone, i.e. a description of the preferred orientations of its architectural elements. Classical techniques are based on the use of a symmetrical second-order tensor, implying orthotropic symmetry within the bone, which is not always the case. In order to properly characterize the architectural anisotropy of cancellous bone samples, this paper presents a new approach for computing the preferred orientations of CT-scanned samples. This method is based on the skeletonization of surfaces reconstructed from binarized CT images, on the use of a projector on the directing vectors of the skeletal rods and finally on the 3D surface analysis of the distribution of the orientations and lengths of the rods. The method introduced has made it possible to obtain the preferred directions of 52 samples of bovine cancellous bone and to show that, in general, these directions are not orthogonal, refuting the oversimplification of orthotropic models.

The chapter defines the modalities of representation and description of the shape of homogeneous regions obtained from the segmentation process. A region can be represented by the pixels of its boundary (external representation) or represented as an aggregation of its pixels (internal representation). Various methods of external representation are presented (based on chain code, polygonal approximation, contour approximation with curved segments, signature, convex hull, skeletonization) and internal (based on texture, color, gray level of the pixels of the region) to objectively characterize the shape of an object. In addition, the shape measurements of a region that are useful for the object recognition process are presented. These shape measurements are extracted using an internal or external representation of the region, based on statistical moments, Fourier descriptors, on moments based on orthogonal bases functions or based on geometric measurements (area, perimeter, compactness, elongation, eccentricity, axes of inertia, rectangularity, etc.). Finally, various shape descriptors invariant to the rotation, translation and scale are described through a normalization process.

Current treatment method for acne and wrinkles requires medical staff to operate with medical injection devices, which requires a lot of time and manpower. In this paper, a method to automate acne and wrinkles treatment using robotic manipulator with two RGB-D camera system is proposed. Through experiments with a UNIVERSAL ROBOT UR5 equipped with the medical injector, it is demonstrated that the proposed approach enables the camera on UR5 to detect and track acne and wrinkles on the flat head face mannequin. Using intrinsic and extrinsic matrix of the RGB-D camera, the coordinates of the targets can be automatically identified with precision. Such targets can be wrinkles or acne which can be detected by skeletonization or template matching methods in the camera system. The system will match the targets and plan the trajectory for the manipulator to follow. The method used in this research to estimate the normal vector of target is PCA (Principal Component Analysis). By trajectory planning, the angles of the manipulator joints can be determined through UR5 inverse kinematics solutions. With such method and optimizing the redundancy of freedom to reduce the motion time by Hill Climbing algorithm, the manipulator is able to move vertically along the skin and the injector can be controlled to aim to the target with precision. To ensure safety, the other camera is fixed on table to monitor working area of UR5 to avoid the manipulator end-effector hitting obstacles. This process enables the possibility of automated medical treatment.

The present work applied different image processing techniques like green component image, background estimation and image skeletonization on the subject’s fundus images. Statistical methods like fractal dimensions, neighbourhood concept was used to distinguish between normal and abnormal fundus images in subjects (n = 45). The results show that, in normal fundus images the vein structures were clearly visible, while in the fundoscopic positive images, the vein structures were totally absent. In fundoscopic negative images the visible vein structures are observed to be thick and coiled up. No significant changes were found in Fractal Dimension (FD) values among the subjects. Neighbourhood pixels (NP) values were found to be 45 ± 0.74 (mean ± S.D.) for normal subjects, 34 ± 1.01 for fundoscopic positive subjects, 20.47 ± 0.49 for fundoscopic negative subjects. The results of this work validated the skeletonized images and support the strength of diagnosis with the help of accurate figures.

This paper presents a probabilistic superpixel-based method for detecting road crack networks. The proposed method includes the techniques of skeletonization and end-growing at the superpixel level, which lend to the extraction of slender crack features from road images. Probabilistic crack pixel refinement is implemented, followed by geometry filters and binary crack cleaning operations, with the end goal of presenting cracks in their simplest form for further high-level characterization. The performance study used to characterize this crack detection algorithm was not constrained by crack type, pavement type, or even image resolution. This approach boasts a median pixel-wise distance error rate of less than one pixel, and for a 100-image dataset, the average detected crack length was within 18% of the ground truth crack length.

Thinning is an iterative object reduction: border points that satisfy some topological and geometric constraints are deleted until stability is reached. If a border point is not deleted in an iteration, conventional implementations take it into consideration again in the next step. With the help of the concepts of a 2D-simplifier point and a weak-3D-simplifier point, rechecking of some ‘survival’ points is not needed. In this work an implementation scheme is reported for sequential thinning algorithms, and it is shown that the proposed method can be twice as fast as the conventional approach in the 2D case.

In order to realize the design of vision-based cement crack repair robot, it is necessary to accurately recognize and extract features of cracks. In this paper, three kinds of typical crack are selected to study, which are fine crack, reticulated crack and dark crack. Firstly, image filtering and image enhancement are used to pre-process the collected image to reduce the influence of noise on detection and enhance the contrast between image background and crack area. Then, the multi-scale morphological operation is applied to extract the fracture edge features effectively. The experimental results show that the proposed edge regions are obviously different from the background regions. Furthermore, by calculating and selecting the area of the largest connected area, the noise can be eliminated to the greatest extent. Finally, the traditional skeleton extraction algorithm is improved to eliminate the number of burrs in the traditional skeleton algorithm. By remapping the cracks images to color images, it can be found that the crack recognition and skeleton extraction meet the requirements, which can provide corresponding technical support for the navigation design of the crack repair robot.

Graph Convolutional Networks (GCNs) are a novel and powerful method for dealing with non-Euclidean data, while Convolutional Neural Networks (CNNs) can learn features from Euclidean data such as images. In this work, we propose a novel method to combine CNNs with GCNs (CNN-GCN), that can consider both Euclidean and non-Euclidean features and can be trained end-to-end. We applied this method to separate the pulmonary vascular trees into arteries and veins (A/V). Chest CT scans were pre-processed by vessel segmentation and skeletonization, from which a graph was constructed: voxels on the skeletons resulting in a vertex set and their connections in an adjacency matrix. 3D patches centered around each vertex were extracted from the CT scans, oriented perpendicularly to the vessel. The proposed CNN-GCN classifier was trained and applied on the constructed vessel graphs, where each node is then labeled as artery or vein. The proposed method was trained and validated on data from one hospital (11 patient, 22 lungs), and tested on independent data from a different hospital (10 patients, 10 lungs). A baseline CNN method and human observer performance were used for comparison. The CNN-GCN method obtained a median accuracy of 0.773 (0.738) in the validation (test) set, compared to a median accuracy of 0.817 by the observers, and 0.727 (0.693) by the CNN. In conclusion, the proposed CNN-GCN method combines local image information with graph connectivity information, improving pulmonary A/V separation over a baseline CNN method, approaching the performance of human observers.

We propose a novel Voronoi Diagram based skeletonization algorithm that produces non-centered skeletons. The first strategy considers utilizing Elliptical Line Voronoi Diagrams with varied density based sampling of the polygonal shapes. The second strategy applies a weighting scheme on Elliptical Line Voronoi Diagrams and Line Voronoi Diagrams. The proposed skeletonization algorithm uses precomputed distance fields and basic element-wise operations, thus can be easily adapted for parallel execution. Non-centered Voronoi Skeletons give a representation that is more similar to real world skeletons and retain many of the desirable properties of skeletons.