Artificial Images Research Articles

Semi-Supervised Natural Face De-Occlusion

Occlusions are often present in face images in the wild, e.g., under video surveillance and forensic scenarios. Existing face de-occlusion methods are limited as they require the knowledge of an occlusion mask. To overcome this limitation, we propose in this paper a new generative adversarial network (named OA-GAN) for natural face de-occlusion without an occlusion mask, enabled by learning in a semi-supervised fashion using (i) paired images with known masks of artificial occlusions and (ii) natural images without occlusion masks. The generator of our approach first predicts an occlusion mask, which is used for filtering the feature maps of the input image as a semantic cue for de-occlusion. The filtered feature maps are then used for face completion to recover a non-occluded face image. The initial occlusion mask prediction might not be accurate enough, but it gradually converges to the accurate one because of the adversarial loss we use to perceive which regions in a face image need to be recovered. The discriminator of our approach consists of an adversarial loss, distinguishing the recovered face images from natural face images, and an attribute preserving loss, ensuring that the face image after de-occlusion can retain the attributes of the input face image. Experimental evaluations on the widely used CelebA dataset and a dataset with natural occlusions we collected show that the proposed approach can outperform the state of the art methods in natural face de-occlusion.

Content Based Image Retrieval on Natural and Artificial Images

Data or information now is not only presented in text, but can also be presented in other forms for example in the form of images. Still image data and moving picture require storage which is much bigger than the text data. As a consequence, we need a method that can speed up the process of indexing and retrieving digital images as well as to reduce storage needed. So increasing data in the form of digital images does not reduce the speed and performance of computers and image data transfer in the internet. In this paper, we try to explore the JPEG image which is currently considered as an international standard for images in WWW. Most of the images currently stored in a media that we use frequently or on the internet are in JPEG format. This JPEG format has many advantages compared to other formats one of the advantage of JPEG is the size smaller compared to other such as PNG, and TIF. Therefore, JPEG has a very important role in saving storage without reducing quality of image itself. The research was carried out to determine the effectiveness and speed of retrieving natural images and artificial images using the CBIR method, research using no less than 5,000 images, and applying queries from 50 to 80. The results showed than in terms of effectiveness, images artificially better than natural images while natural speed is better. On the hand, speed of retrieving of natural images much better compared to the artificial images.

Simultaneous super-resolution and contrast synthesis of routine clinical magnetic resonance images of the knee for improving automatic segmentation of joint cartilage: data from the Osteoarthritis Initiative.

High resolution three-dimensional (3D) magnetic resonance (MR) images are well suited for automated cartilage segmentation in the human knee joint. However, volumetric scans such as 3D Double-Echo Steady-State (DESS) images are not routinely acquired in clinical practice which limits opportunities for reliable cartilage segmentation using (fully) automated algorithms. In this work, a method for generating synthetic 3D MR (syn3D-DESS) images with better contrast and higher spatial resolution from routine, low resolution, two-dimensional (2D) Turbo-Spin Echo (TSE) clinical knee scans is proposed. A UNet convolutional neural network is employed for synthesizing enhanced artificial MR images suitable for automated knee cartilage segmentation. Training of the model was performed on a large, publically available dataset from the OAI, consisting of 578 MR examinations of knee joints from 102 healthy individuals and patients with knee osteoarthritis. The generated synthetic images have higher spatial resolution and better tissue contrast than the original 2D TSE, which allow high quality automated 3D segmentations of the cartilage. The proposed approach was evaluated on a separate set of MR images from 88 subjects with manual cartilage segmentations. It provided a significant improvement in automated segmentation of knee cartilages when using the syn3D-DESS images compared to the original 2D TSE images. The proposed method can successfully synthesize 3D DESS images from 2D TSE images to provide images suitable for automated cartilage segmentation.

Evaluation of Defect Detection in Aluminium, CFRP and Epoxy Resin Plates Using Non-contact Air-Coupled Ultrasonic Waves

In the field of aerospace materials and structures, water coupled ultrasonic testing was not desirable due to changes in properties and contamination reason. Therefore, in the present study non-contact air coupled ultrasonic transducers were proposed as an alternative examination for such applications. This present study describes our practical knowledge of using piezoelectric air-coupled transducers for visualization and defect identification of aluminium, CFRP and epoxy-resin specimens. Further, the effectiveness of the system is verified by experiments conducted in the inspection of artificial defects and 2D images was reconstructed through scanning of specific specimen surface. Results show that, all the created defects were identified successfully using C-scan technique. The measured defect values were compared with experimental defect values and shows good agreement. The 1 MHz piezoelectric transducer was the best for all the three specimens which is used in the present study because of the material structure being evaluated. The bulk wave piezoelectric transducer was identified as extensive pulse attenuation by the structure of the aluminium, CFRP and epoxy resin.

A visual framework to create photorealistic retinal vessels for diagnosis purposes

The methods developed in recent years for synthesising an ocular fundus can be been divided into two main categories. The first category of methods involves the development of an anatomical model of the eye, where artificial images are generated using appropriate parameters for modelling the vascular networks and fundus. The second type of method has been made possible by the development of deep learning techniques and improvements in the performance of hardware (especially graphics cards equipped with a large number of cores). The methodology proposed here to produce high-resolution synthetic fundus images is intended to be an alternative to the increasingly widespread use of generative adversarial networks to overcome the problems that arise in producing slightly modified versions of the same real images. This will allow the simulation of pathologies and the prediction of eye-related diseases. The proposed approach is based on the principle of least action and correctly places the vessels on the simulated eye fundus without using real morphometric information. An a posteriori analysis of the average characteristics such as the size, length, bifurcations, and endpoint positioning confirmed the substantial accuracy of the proposed approach compared to real data. A graphical user interface allows the user to make any changes in real time by controlling the positions of control points.

Assessing the Big Five personality traits using real-life static facial images

There is ample evidence that morphological and social cues in a human face provide signals of human personality and behaviour. Previous studies have discovered associations between the features of artificial composite facial images and attributions of personality traits by human experts. We present new findings demonstrating the statistically significant prediction of a wider set of personality features (all the Big Five personality traits) for both men and women using real-life static facial images. Volunteer participants (N = 12,447) provided their face photographs (31,367 images) and completed a self-report measure of the Big Five traits. We trained a cascade of artificial neural networks (ANNs) on a large labelled dataset to predict self-reported Big Five scores. The highest correlations between observed and predicted personality scores were found for conscientiousness (0.360 for men and 0.335 for women) and the mean effect size was 0.243, exceeding the results obtained in prior studies using ‘selfies’. The findings strongly support the possibility of predicting multidimensional personality profiles from static facial images using ANNs trained on large labelled datasets. Future research could investigate the relative contribution of morphological features of the face and other characteristics of facial images to predicting personality.

Development of phased array ultrasonic system for detecting rail cracks

In this study, a phased array ultrasonic system for detecting rail cracks (PAUSR) was developed which consists of two phased array (PA) ultrasonic transducers, a water tank, a display monitor, a battery, a commercial 64 channel PA board, and its control software. To accomplish this, the acoustic fields and crack detection images of newly developed PA ultrasonic transducers were simulated and analyzed by the CIVA (CIVA 2016, NDE CIVA, USA) software. The major design factors for the PAUSR is the capability of evaluating crack size (over 2 mm), generating proper acoustic fields in the rail and easy and safe handling for operators. For the performance test of the designed PA ultrasonic transducer, Non-destructive Characterization and Verification of Ultrasonic Phased Array Equipment, Part 2: Probes (EN ISO 18563-2:2017) was conducted. In addition, artificial crack detection images were simulated by the CIVA software, and an experiment under the same CIVA simulation conditions was also performed for comparison purposes. For the last verification, natural defective specimens with bolt-hole star-cracking, longitudinal vertical cracking and welding defects were scanned by the developed PAUSR, and its results were analyzed and compared with the radiographic testing (RT) and CIVA simulation outcomes.

Point-by-point feature extraction of artificial intelligence images based on the Internet of Things

In the context of the rapid development of artificial intelligence in the Internet of Things, the establishment of the Internet of Things can promote rapid progress in the field of artificial intelligence. Traditional image detection methods use wavelet energy algorithms to divide background and edge noise, with poor resolution and low image detection accuracy. A series of problems such as slow detection speed and lack of image depth analysis exists. Aiming at the disadvantages of traditional methods, this study proposes the design of an artificial intelligence image detection system based on the Internet of Things, and uses intelligent artificial pixel feature collection technology to extract point-by-point feature of the image. This paper introducing artificial intelligence learning algorithms to the wheel detection in the workshop under the Internet of Things system, which can not only solve the problem of poor feature anti-interference and poor robustness in the traditional method, but also has important significance for the secondary development of the wheel detection system. The neural network can be used to classify wheel images while integrating other detection requirements, such as wheel defect detection and wheel number identification. The rich data resources and processing capabilities of the Internet of Things to perform feature analysis and feedback on the collected image pixels are utilized. The artificial intelligence of image synthesis module performs image conversion processing on the signal processes the feedback signal. The analysis results can complete image detection and complete artificial intelligence images. Through simulation experiments, it is proved that the design of artificial intelligence image detection system based on the Internet of Things has the advantages of high image detection rate, high recognition accuracy, stable operation, and efficient processing. The design idea has good application value.

On the use of a cascaded convolutional neural network for three-dimensional flow measurements using astigmatic PTV

Many applications in chemistry, biology and medicine use microfluidic devices to separate, detect and analyze samples on a miniaturized size-level. Fluid flows evolving in channels of only several tens to hundreds of micrometers in size are often of a 3D nature, affecting the tailored transport of cells and particles. To analyze flow phenomena and local distributions of particles within those channels, astigmatic particle tracking velocimetry (APTV) has become a valuable tool, on condition that basic requirements like low optical aberrations and particles with a very narrow size distribution are fulfilled. Making use of the progress made in the field of machine vision, deep neural networks may help to overcome these limiting requirements, opening new fields of applications for APTV and allowing them to be used by nonexpert users. To qualify the use of a cascaded deep convolutional neural network (CNN) for particle detection and position regression, a detailed investigation was carried out starting from artificial particle images with known ground truth to real flow measurements inside a microchannel, using particles with uni- and bimodal size distributions. In the case of monodisperse particles, the mean absolute error and standard deviation of particle depth-position of less than and about were determined, employing the deep neural network and the classical evaluation method based on the minimum Euclidean distance approach. While these values apply to all particle size distributions using the neural network, they continuously increase towards the margins of the measurement volume of about one order of magnitude for the classical method, if nonmonodisperse particles are used. Nevertheless, limiting the depth of measurement volume in between the two focal points of APTV, reliable flow measurements with low uncertainty are also possible with the classical evaluation method and polydisperse tracer particles. The results of the flow measurements presented herein confirm this finding. The source code of the deep neural network used here is available on https://github.com/SECSY-Group/DNN-APTV.

Intuitionistic based segmentation of thyroid nodules in ultrasound images

Accurate delineation of thyroid nodules in ultrasound images is vital for computer-aided diagnosis. Most segmentation methods are semi-automated for thyroid nodules and require manual intervention, which increases the processing time and errors. We propose an automated intuitionistic fuzzy active contour method (IFACM) that integrates intuitionistic fuzzy clustering with an active contour for thyroid nodule segmentation using ultrasound images. Intuitionistic fuzzy clustering is used for the initialization of an active contour and estimation of the parameters required to automatically control the curve evolution. The IFACM was tested extensively on both artificial and real ultrasound images. The IFACM obtained a higher value of true positive (95.1% ± 2.86%), overlap metric (93.1 ± 2.95%), and dice coefficient (90.90 ± 3.08), indicating that the boundary delineated by the IFACM fits best to true nodules. Moreover, it obtained a lower value of false positive (04.1% ± 3.24%) and Hausdorff distance (0.50 ± 0.21 in pixels), further verifying the higher similarity of shape and boundary, respectively. According to the significance test, the results of the proposed method were more significant than those of the other segmentation methods. The main benefit of the IFACM is the automatic identification of nodules on the basis of image characteristics, which eliminates manual intervention. In all the experiments, all initial contours were automatically defined closer to the boundaries of the nodule, which is a benefit of the IFACM. Moreover, this method can segment multiple nodules in a single image efficiently.

Expression of Fractals Through Neural Network Functions

To help understand the underlying mechanisms of neural networks (NNs), several groups have studied the number of linear regions $\ell $ of piecewise linear (PwL) functions, generated by deep neural networks (DNN). In particular, they showed that $\ell $ can grow exponentially with the number of network parameters $p$ , a property often used to explain the advantages of deep over shallow NNs. Nonetheless, a dimension argument shows that DNNs cannot generate all PwL functions with $\ell $ linear regions when $\ell > p$ . It is thus natural to seek to characterize specific families of functions with $\ell > p$ linear regions that can be constructed by DNNs. Iterated Function Systems (IFS) recursively construct a sequence of PwL functions $F_{k}$ with a number of linear regions which is exponential in $k$ . We show that $F_{k}$ can be generated by a NN using only $\mathcal {O}(k)$ parameters. IFS are used extensively to generate natural-looking landscape textures in artificial images as well as for compression of natural images. The surprisingly good performance of this compression suggests that human visual system may lock in on self-similarities. The combination of this phenomenon with the capacity of DNNs to efficiently approximate IFS may contribute to the success of DNNs, particularly striking for image processing tasks.

Classification of hyperspectral images of the interior of fruits and vegetables using a 2D convolutional neuronal network

This work evaluates the performance of a Deep Learning technique for classification of challenging hyperspectral images of the interior of fruits and vegetables when they are combined. Some of these samples have low contrast, similar colour features, and their skins or characteristic shapes are lost when cut to expose their interiors. We implemented a two-dimensional convolutional neural network for this classification task and compared their results against the technique of support vector machines. We randomly selected a group of 13 hyperspectral images from a public database containing information of the interior of 42 fruits and vegetables. Using parts of these 13 selected images, we constructed three artificial hyperspectral images merging these parts differently. We applied the two proposed techniques over the three of them. The comparison of the classification results shows that the two-dimensional convolutional neural network over-performs the support vector machine in all three composite images. The two-dimensional convolutional neural network exceeded 98% classification accuracy in all of them. These results show that the two-dimensional convolutional network benefits from the spatial and spectral data in the images obtaining proper levels of classification even in samples mixed in complex contexts, as it can occur in the food or pharmaceutical industries.

Data augmentation for handwritten digit recognition using generative adversarial networks

Supervised learning techniques require labeled examples that can be time consuming to obtain. In particular, deep learning approaches, where all the feature extraction stages are learned within the artificial neural network, require a large number of labeled examples to train the model. Various data augmentation techniques can be performed to overcome this issue by taking advantage of known variations that have no impact on the label of an example. Typical solutions in computer vision and document analysis and recognition are based on geometric transformations (e.g. shift and rotation) and random elastic deformations of the original training examples. In this paper, we consider Generative Adversarial Networks (GAN), a technique that does not require prior knowledge of the possible variabilities that exist across examples to create novel artificial examples. In the case of a training dataset with a low number of labeled examples, which are described in a high dimensional space, the classifier may generalize poorly. Therefore, we aim at enriching databases of images or signals for improving the classifier performance by designing a GAN for creating artificial images. While adding more images through a GAN can help, the extent to which it will help is unknown, and it may degrade the performance if too many artificial images are added. The approach is tested on four datasets on handwritten digits (Latin, Bangla, Devanagri, and Oriya). The accuracy for each dataset shows that the addition of GAN generated images in the training dataset provides an improvement of the accuracy. However, the results suggest that the addition of too many GAN generated images deteriorates the performance.

Improving watershed-based pore-network extraction method using maximum inscribed ball pore-body positioning

Pore-scale modeling is a rapidly evolving area of research because modeling multiphase flow directly on 3D pore geometries is of utmost importance in wide variety of research areas, including oil and gas development, hydrology and material sciences. Although there are numerous methods to model flow, only so-called pore-network models are computationally effective enough to perform simulations in large modeling domains, and they are orders of magnitude faster than direct modeling approaches. However, pore-network models require a simplification of the 3D pore geometry to perform simulations, which are usually referred to as pore-network extraction. Such extraction poses a separate problem because it must provide an accurate description of the pore space geometry and topology. Different methods have been proposed in the literature. Recently, watershed-based approaches have been popular due to their effectiveness in working with porous media images of any porosity. A watershed algorithm requires seed placement to segment the space into distinct pores. We propose a hybrid algorithm combining the power of watersheds in finding intersections between pores with the advantages of the maximum inscribed ball technique, which is very effective in finding pore centers. We rigorously verify and test our novel methodology on artificial and X-ray microtomography images of wide variety of porous materials: sphere packings, carbonate, soil, ceramic and sandstone samples. Comparison against a purely watershed-based method and results based solely on the maximum inscribed balls–based method (in terms of pore/throat total number, pore size distributions and connection statistics, and multiphase flow properties including capillary curves and relative permeabilities) revealed the accuracy of our novel technique, consistency with existing classical techniques and great potential in analysing 3D pore images of any complexity. On the other hand, comparison of extracted pore-network topology (as based on Euler number) revealed significant differences between different methodologies, which is rather surprising considering the similarities in two-phase flow properties. While analysing permeability results, we also compared two popular pore-throat partitioning models and advocated in favor of the weight model usually utilized within watershed-based extracted pore networks. Our results illuminate problems in current pore-network models and outline some potential ways to improve their accuracy in future research.

An image processing algorithm for the measurement of multiphase bubbly flow using predictor-corrector method

Optical photography and image analysis technology provide a non-intrusive and efficient research tool for the experimental research of multiphase bubbly flow. This paper presented a multi-frame image processing algorithm for measuring size and velocity of bubbles more accurately with bubble shadow images, especially in bubbly plume with serious overlapping in the bubble image. The raw images are the accurate recording of bubble shadow by high frequency CCD camera. In the shadow image, some of bubbles are isolated and others may overlap resulting in a complex geometric shape. It is easy to obtain the shape and centroid of isolated bubbles. But for the overlapping bubbles, additional effort should be made to detect each one from a cluster. And some of them are highly and completely overlapped, and may be lost during tracking. The main conception of the algorithm is based on the continuity of bubble movement to retrieve those bubbles. In one frame, a bubble which is highly or completely overlapped by others may appear alone in another frame. With the information of previous frames, the centroid and size of highly overlapping bubbles are predicted and then corrected according internal cores and edge information of bubble shadow. The algorithm is tested by artificial bubble images and further applied to the experiment of bubbly plume. It is proved that the algorithm performs well in recognition of bubbles and especially in determination of the size and velocity of bubbles in a cluster. For the artificial bubble images, the proposed algorithm successfully captures all bubbles including an inner bubble which cannot be detected in a single frame. For the experiment of bubbly plume, which is conducted in a tank with a 1 cm diameter nozzle at bottom and with the superficial gas velocity ranging from 31.8 to 53.1 mm/s, the effectiveness of the proposed method also reaches 95% by random sampling. For the case that more than half of bubbles are overlapped, the new algorithm can improve the recognition rate from 4% to 6% comparing with the primary algorithm without predictor-corrector method.

Investigation of the emit source coordinates estimation accuracy using reflections from the ground clutter

The issues of improving the accuracy of determining the coordinates of a radio source by a single-position method due to the joint processing of multiple signals reflected from local objects are considered. The electronic terrain map was obtained on the basis of artificial images of a two-dimensional point-like reconstruction flow. The results of computer simulation are presented.

DCE-MRI pharmacokinetic parameter maps for cervical carcinoma prediction

Pharmacokinetic parameters estimated from dynamic contrast-enhanced magnetic resonance imaging (DCE-MRI) time course data enable the physio-biological interpretation of tissue angiogenesis. This study aims to develop machine learning approaches for cervical carcinoma prediction based on pharmacokinetic parameters. The performance of individual parameters was assessed in terms of their efficacy in differentiating cancerous tissue from normal cervix tissue. The effect of combining parameters was evaluated using the following two approaches: the first approach was based on support vector machines (SVMs) to combine the parameters from one pharmacokinetic model or across several models; the second approach was based on a novel method called APITL (artificial pharmacokinetic images for transfer learning), which was designed to fully utilize the comprehensive pharmacokinetic information acquired from DCE-MRI data. A “winner-takes-all” strategy was employed to consolidate the slice-wise prediction into subject-wise prediction. Experiments were carried out with a dataset comprising 36 patients with cervical cancer and 17 healthy subjects. The results demonstrated that parameter Ve, representing volume fraction of the extracellular extravascular space (EES), attained high discriminative power regardless of the pharmacokinetic model used for estimation. An approximately 10% improvement in the accuracy was achieved with the SVM approach. The APITL method further outperformed SVM and attained a subject-wise prediction accuracy of 94.3%. Our experiment demonstrated that APITL could predict cervical carcinoma with high accuracy and had potential in clinical applications.

Overhaul of Three-D Image Based on Metamorphosis and Fabric Growing of Two-D Images

Shape is a critical physical property of normal and artificial three-D images that describes their outside appearances. Understanding differences among shapes and displaying the inconstancy inside and outside the shape classes are considered for shape analysis, and are the major issues in numerous applications, from normal image visualization to medical imaging. During diagnosis in medical image processing it is impossible to analyze the diseased areas some time from three-D images. So for the purpose of diagnosing the diseased areas of three-D image, medical experts need two-D images. This paper addresses the overhaul of three dimensional models from two-D images. In the initial step the image is segmented using level set method. Later segmented image is extracted and registered for overhaul of three dimensional images using metamorphosis and fabric growing methods. The practical result shows the implementation of the suggested method.

Behavioral responses of zebrafish (Danio rerio) to different 2D non-moving stimuli

Paper description: The behavioral response to social and sexual 2D non-moving images representing conspecifics were examined. Sexually mature fish underwent non-invasive behavioral tests in a cross-maze. Artificial non-moving 2D images were created using photoshop graphics software. We observed significant sex-related differences in zebrafish responses to fixed artificial stimuli. Females showed a statistically significant preference for 2D images of phenotypically less similar individuals. These results improve knowledge on zebrafish visual perception regarding shoaling and mate choice. The paper offers an alternative to use of animals in experimental research. Abstract: Zebrafish ( Danio rerio ) is considered an experimental model organism with numerous applications in different fields. The mechanisms underlying social and reproductive preferences are complex. Most of the literature on conspecific interactions focus on the chemical communication mediated by scattered signals in the aquatic medium. The aim of this research was to evaluate the behavioral responses induced in the tested animals by artificial social stimuli. We used 20 sexually mature individuals that underwent several non-invasive behavioral tests in a cross-maze. Artificial non-moving 2D images were created for this study, using photoshop graphics software. Ethological measurements were conducted using video-tracking software (EthoVision XT). The findings showed significant sex-related differences in zebrafish responses to fixed artificial stimuli. In particular, females shoaled more with 2D non-moving images representing conspecifics phenotypically more distant (i.e. color) from them. Integrating dimension and number of individual, one large stimulus was preferred over three small stimuli in terms of cumulative and mean duration for males, and only in terms of cumulative duration for females. We concluded that 2D images representing conspecifics, even without movements, can induce a behavioral response in this species. https://doi.org/10.2298/ABS190823065L Received: August 23, 2019; Revised: November 10, 2019; Accepted: November 12, 2019; Published online: December 11, 2019 How to cite this article: Lenzi C, Grasso C, Nicoara M, Savuca A, Ciobica A, Plavan GI, Strungaru SA. Behavioral responses of zebrafish ( Danio rerio ) to different 2D non-moving stimuli. Arch Biol Sci. 2020;72(1):45-52.

Design of Chinese Character Coded Targets for Feature Point Recognition Under Motion-Blur Effect

Coded targets have been widely used as a type of active visual feature points in fields such as close-range photogrammetry, robot navigation, 3D reconstruction, and augmented reality. However, coded targets in degraded images, such as images with motion blur effect, are hard to recognize. To this end, a set of novel Chinese character coded targets (CCTs) is designed and tested. A CCT is a square visual marker, shown as a relative small circular feature overlaid in the middle of a square Chinese character. A white circular ring is embedded in a black circle concentrically to serve as the circular feature, which facilitates extraction of the center point of the marker for localization. Whereas the distinctive peripheral Chinese character appearance of each CCT is utilized for identification. By synthesizing simulated CCTs with different degrees of motion blur and various postures with the real background images, a Faster Region-based Convolutional Neural Network (Faster R-CNN) is trained to locate and recognize the CCTs in motion blurred images. Experimental results on both artificial and actual motion blurred images demonstrate the superiorities of the designed CCTs as well as the proposed localization and recognition pipeline.