X-ray testing of complex objects -such as luggage screening at airports-is usually performed manu... more X-ray testing of complex objects -such as luggage screening at airports-is usually performed manually. This is not always effective, since it depends strongly on the pose of the objects of interest (target objects), occlusion and human capabilities as well. Additionally, certain target objects are difficult to be detected using only one viewpoint. For this reason, we developed an active X-ray testing framework that is able to adequate the viewpoint of the target object in order to obtain better X-ray images to analyze. The key idea of our method is to adapt automatically the viewpoint of the X-ray images in order to project the target object in poses where the detection performance should be higher. Thus, the detection inside of complex objects can be performed in a more effective way. Using a robotic arm and a semi-automatic manipulator system, the robustness and reliability of the method have been verified in the automated detection of razor blades located inside of nine different objects showing promising preliminary results: in 130 experiments we were able to detect 115 times the razor blade with 10 false alarms, achieving recall of 89% and precision of 92%.
In countries where fish is often consumed, fish bones are some of the most frequently ingested fo... more In countries where fish is often consumed, fish bones are some of the most frequently ingested foreign bodies encountered in foods. In the production of fish fillets, fish bone detection is performed by human inspection using their sense of touch and vision which can lead to misclassification. Effective detection of fish bones in the quality control process would help avoid this problem. For this reason, an X-ray machine vision approach to automatically detect fish bones in fish fillets was developed. This paper describes our approach and the corresponding experiments with salmon and trout fillets. In the experiments, salmon X-ray images using 10×10 pixels detection windows and 24 intensity features (selected from 279 features) were analyzed. The methodology was validated using representative fish bones and trouts provided by a salmon industry and yielded a detection performance of 99%. We believe that the proposed approach opens new possibilities in the field of automated visual inspection of salmon, trout and other similar fish.
This paper describes the main activities and achievements of our research group on Machine Intell... more This paper describes the main activities and achievements of our research group on Machine Intelligence and Robotics (Grima) at the Computer Science Department, Pontificia Universidad Catolica de Chile (PUC). Since 2002, we have been developing an active research in the area of indoor autonomous social robots. Our main focus has been the cognitive side of Robotics, where we have developed algorithms for autonomous navigation using wheeled robots, scene recognition using vision and 3D range sensors, and social behaviors using Markov Decision Processes, among others. As a distinguishing feature, in our research we have followed a probabilistic approach, deeply rooted in machine learning and Bayesian statistical techniques. Among our main achievements are an increasing list of publications in main Robotics conference and journals, and the consolidation of a research group with more than 25 people among full-time professors, visiting researchers, and graduate students.
The classic image processing method for flaw detection uses one image of the scene, or multiple i... more The classic image processing method for flaw detection uses one image of the scene, or multiple images without correspondences between them. To improve this scheme, automated inspection using multiple views has been developed in recent years. This strategy's key idea is to consider as real flaws those regions that can be tracked in a sequence of multiple images because they are located in positions dictated by geometric conditions. In contrast, false alarms (or noise) can be successfully eliminated in this manner, since they do not appear in the predicted places in the following images, and thus cannot be tracked. This paper presents a method to inspect aluminum wheels using images taken from different positions by using a method called automatic multiple view inspection (AMVI). Our method can be applied to uncalibrated image sequences, therefore it is not necessary to determine optical and geometric parameters normally present in the calibrated systems. In addition, to improve the performance, we designed a false alarm reduction method in two and three views called Intermediate Classifier Block (ICB). The ICB method takes advantage of the classifier ensemble methodology by making use of feature analysis in multiple views. Using this method, real flaws can be detected with high precision while most false alarms can be discriminated.
Substantial research has been performed on automated detection and classification of welding defe... more Substantial research has been performed on automated detection and classification of welding defects in continuous welds using X-ray imaging. Typically, the detection follows a pattern recognition schema (segmentation, feature extraction and classification). In computer vision community, however, many object detection and classification problems, like face and human detection, have been recently solvedwithout segmentation-using sliding-windows and novel features like local binary patterns extracted from saliency maps. For this reason, we propose in this paper the use of sliding-windows with the mentioned features to perform automatically the automated detection of welding defects. In the experiments, we analyzed 5000 detection windows (24x24 pixels) and 572 intensity features from 10 representative X-ray images. Cross validation yielded a detection performance of 94% using a support vector machine classifier with only 14 selected features. The method was implemented and tested on real X-ray images showing high effectiveness. We believe that the proposed approach opens new possibilities in the field of automated detection of welding defects.
Deep-fat frying is a unit operation which develops unique sensorial attributes in foods. For inst... more Deep-fat frying is a unit operation which develops unique sensorial attributes in foods. For instance, texture is the principal quality parameter of tortilla and corn chips. On the other hand, computer vision is a useful tool for quality evaluation and prediction of some physical properties in different either raw or processed foods. The objective of this research was to characterize corn and tortilla chips by using computer vision, and to build proper mathematical models which permit to predict mechanical properties of these chips (maximum force, such as hardness, and distance to maximum force, such as toughness) by using chromatic features extracted from their corresponding digital images. Corn and tortilla chips (thickness of 2 mm; diameter of 37 mm) were made from masa of maize and fried at constant oil temperatures of 160, 175, and 190°C. A high linear correlation (R 2 >0.9400) was obtained between mechanical properties and some image features (Hu, Fourier, and Haralick moments). Cross-validation technique demonstrated the repeatability and good performance (>90%) of the models tested, indicating that can be used to predict the textural properties of the tortilla and corn chips by using selected features extracted from their digital images, without the necessity of measuring them in a texture analyzer.
X-ray testing of complex objects -such as luggage screening at airports-is usually performed manu... more X-ray testing of complex objects -such as luggage screening at airports-is usually performed manually. This is not always effective, since it depends strongly on the pose of the objects of interest (target objects), occlusion and human capabilities as well. Additionally, certain target objects are difficult to be detected using only one viewpoint. For this reason, we developed an active X-ray testing framework that is able to adequate the viewpoint of the target object in order to obtain better X-ray images to analyze. The key idea of our method is to adapt automatically the viewpoint of the X-ray images in order to project the target object in poses where the detection performance should be higher. Thus, the detection inside of complex objects can be performed in a more effective way. Using a robotic arm and a semi-automatic manipulator system, the robustness and reliability of the method have been verified in the automated detection of razor blades located inside of nine different objects showing promising preliminary results: in 130 experiments we were able to detect 115 times the razor blade with 10 false alarms, achieving recall of 89% and precision of 92%.
Considerable research efforts in computer vision applied to food quality evaluation have been dev... more Considerable research efforts in computer vision applied to food quality evaluation have been developed in the last years; however, they have been concentrated on using or developing tailored methods based on visual features that are able to solve a specific task. Nevertheless, today's computer capabilities are giving us new ways to solve complex computer vision problems. In particular, a new paradigm on machine learning techniques has emerged posing the task of recognizing visual patterns as a search problem based on training data and a hypothesis space composed by visual features and suitable classifiers. Furthermore, now we are able to extract, process, and test in the same time more image features and classifiers than before. Thus, we propose a general framework that designs a computer vision system automatically, i.e., it findswithout human interaction-the features and the clas-sifiers for a given application avoiding the classical trial and error framework commonly used by human designers. The key idea of the proposed framework is to select-automatically-from a large set of features and a bank of classifiers those features and classifiers that achieve the highest performance. We tested our framework on eight different food quality evaluation problems yielding a classification performance of 95 % or more in every case. The proposed framework was implemented as a Matlab Toolbox available for noncommercial purposes.
In order to reduce the security risk of a commercial aircraft, passengers are not allowed to take... more In order to reduce the security risk of a commercial aircraft, passengers are not allowed to take certain items in carry-on baggage. For this reason, human operators are trained to detect prohibited items using a manually controlled baggage screening process. In this paper, we propose the use of a method based on multiple X-ray views to detect some regular prohibited items with very defined shapes and sizes. The method consists of two steps: 'structure estimation', to obtain a geometric model of the multiple views from the object to be inspected (a baggage), and 'parts detection', to detect the parts of interest (prohibited items). The geometric model is estimated using a structure from motion algorithm. The detection of the parts of interest is performed by an adhoc segmentation algorithm (object dependent) followed by a general tracking algorithm based on geometric and appearance constraints. In order to illustrate the effectiveness of the proposed method, experimental results on detecting regular objects −razor blades and guns− are shown yielding promising results.
This paper presents a new methodology for identifying parts of interest inside of a complex objec... more This paper presents a new methodology for identifying parts of interest inside of a complex object using multiple X-ray views. The proposed method consists of five steps: A) image acquisition, that acquires an image sequence where the parts of the object are captured from different viewpoints; B) geometric model estimation, that establishes a multiple view geometric model used to find the correct correspondence among different views; C) single view detection, that segment potential regions of interest in each view; D) multiple view detection, that matches and tracks potential regions based on similarity and geometrical multiple view constraints; and E) analysis, that analyzes the tracked regions using multiple view information, filtering out false alarms without eliminating existing parts of interest. In order to evaluate the effectiveness of the proposed method, the algorithm was tested on 32 cases (five applications using different segmentation approaches) yielding promising results: precision and recall were 95.7% and 93.9%, respectively. Additionally, the multiple view information obtained from the tracked parts was effectively used for recognition purposes. In our recognition experiments, we obtained an accuracy of 96.5%. Validation experiments show that our approach achieves better performance than other representative methods in the literature.
X-ray testing has been developed for inspection of materials or objects, where the aim is to anal... more X-ray testing has been developed for inspection of materials or objects, where the aim is to analyze -nondestructively-those inner parts that are undetectable to the naked eye. Thus, Xray testing is used to determine if a test object deviates from a given set of specifications. Typical applications are inspection of automotive parts, quality control of welds, screening of baggage, analysis of food products and inspection of cargos. In order to achieve efficient and effective X-ray testing, automated and semi-automated systems based on computer vision algorithms are being developed to execute this task. In this paper, we present a general overview of computer vision approaches that have been used in X-ray testing. In addition, we review some techniques that have been applied in certain relevant applications; and we introduce a public database of X-ray images that can be used for testing and evaluation of image analysis and computer vision algorithms. Finally, we conclude that there are some areas -like casting inspection where automated systems are very effective, and other application areas -such as baggage screening-where human inspection is still used; there are certain application areas -like weld and cargo inspections-where the process is semiautomatic; and there is some research in areas -including food analysis-where processes are beginning to be characterized by the use of X-ray imaging.
In order to reduce the security risk of a commercial aircraft, passengers are not allowed to take... more In order to reduce the security risk of a commercial aircraft, passengers are not allowed to take certain items in their carry-on baggage. For this reason, human operators are trained to detect prohibited items using a manually controlled baggage screening process. The inspection process, however, is highly complex as hazardous items are very difficult to detect when placed in close packed bags, superimposed by other objects, and/or rotated showing an unrecognizable profile. In this paper, we review certain advances achieved by our research group in this field. Our methodology is based on multiple view analysis, because it can be a powerful tool for examining complex objects in cases in which uncertainty can lead to misinterpretation. In our approach, multiple views (taken from fixed points of view, or using an active vision approach in which the best views are automated selected) are analyzed in the detection of regular objects. In order to illustrate the effectiveness of the proposed method, experimental results on recognizing guns, razor blades, pins, clips and springs in baggage inspection are presented achieving around 90% accuracy. We believe that it would be possible to design an automated aid in a target detection task using the proposed algorithm.
Baggage inspection using X-ray screening is a priority task that reduces the risk of crime and te... more Baggage inspection using X-ray screening is a priority task that reduces the risk of crime and terrorist attacks. Manual detection of threat items is tedious because very few bags actually contain threat items and the process requires a high degree of concentration. An automated solution would be a welcome development in this field. We propose a methodology for automatic detection of threat objects using single X-ray images. Our approach is an adaptation of a methodology originally created for recognizing objects in photographs based on Implicit Shape Models. Our detection method uses a visual vocabulary and an occurrence structure generated from a training dataset that contains representative X-ray images of the threat object to be detected. Our method can be applied to single views of grayscale X-ray images obtained using a single energy acquisition system. We tested the effectiveness of our method for the detection of three different threat objects: razor blades, shuriken (ninja stars) and handguns. The testing dataset for each threat object consisted of 200 X-ray images of bags. The true positive and false positive rates (TPR, FPR) are: (0.99, 0.02) for razor blades, (0.97, 0.06) for shuriken and (0.89, 0.17) for handguns. If other representative training datasets were utilized, we believe that our methodology could aid in the detection of other kinds of threat objects.
In this paper, we present a new dataset consisting of 19,407 X-ray images. The images are organiz... more In this paper, we present a new dataset consisting of 19,407 X-ray images. The images are organized in a public database called GDXray that can be used free of charge, but for research and educational purposes only. The database includes five groups of X-ray images: castings, welds, baggage, natural objects and settings. Each group has several series, and each series several X-ray images. Most of the series are annotated or labeled. In such cases, the coordinates of the bounding boxes of the objects of interest or the labels of the images are available in standard text files. The size of GDXray is 3.5 GB and it can be downloaded from our website. We believe that GDXray represents a relevant contribution to the X-ray testing community. On the one hand, students, researchers and engineers can use these X-ray images to develop, test and evaluate image analysis and computer vision algorithms without purchasing expensive X-ray equipment. On the other hand, these images can be used as a benchmark in order to test and compare the performance of different approaches on the same data. Moreover, the database can be used in the training programs of human inspectors.
It is well known that some facial attributes -like soft biometric traits-can increase the perform... more It is well known that some facial attributes -like soft biometric traits-can increase the performance of traditional biometric systems and help recognition based on human descriptions. In addition, other facial attributes, such as facial expressions, can be used in human-computer interfaces, image retrieval, talking heads and human emotion analysis. This paper addresses the problem of automated recognition of facial attributes by proposing a new general approach called Adaptive Sparse Representation of Random Patches (ASR+). The proposed method consists of two stages: In the learning stage, random patches are extracted from representative face images of each class (e.g., in gender recognition -a two-class problem-, images of females/males) in order to construct representative dictionaries. A stop list is used to remove very common words of the dictionaries. In the testing stage, random test patches of the query image are extracted, and for each non-stopped test patch a dictionary is built concatenating the 'best' representative dictionary of each class. Using this adapted dictionary, each non-stopped test patch is classified following the Sparse Representation Classification (SRC) methodology. Finally, the query image is classified by patch voting. Thus, our approach is able to learn a model for each recognition task dealing with a larger degree of variability in ambient lighting, pose, expression, occlusion, face size and distance from the camera. Experiments were carried out on eight face databases in order to recognize facial expression, gender, race, disguise and beard. Results show that ASR+ deals well with unconstrained conditions, outperforming various representative methods in the literature in many complex scenarios.
Minimum Norm Estimation (MNE) is an inverse solution method widely used to reconstruct the source... more Minimum Norm Estimation (MNE) is an inverse solution method widely used to reconstruct the source time series that underlie magnetoencephalography (MEG) data. MNE addresses the ill-posed nature of MEG source estimation through regularization (e.g., Tikhonov regularization). Selecting the best regularization parameter is a critical step. Generally, once set, it is common practice to keep the same coefficient throughout a study. However, it is yet to be known whether the optimal lambda for spectral power analysis of MEG source data coincides with the optimal regularization for source-level oscillatory coupling analysis. We addressed this question via extensive Monte-Carlo simulations of MEG data, where we generated 21,600 configurations of pairs of coupled sources with varying sizes, signal-to-noise ratio (SNR), and coupling strengths. Then, we searched for the Tikhonov regularization coefficients (lambda) that maximize detection performance for (a) power and (b) coherence. For coherence, the optimal lambda was two orders of magnitude smaller than the best lambda for power. Moreover, we found that the spatial extent of the interacting sources and SNR, but not the extent of coupling, were the main parameters affecting the best choice for lambda. Our findings suggest using less regularization when measuring oscillatory coupling compared to power estimation.
In recent years, X-ray screening systems have been used to safeguard environments in which access... more In recent years, X-ray screening systems have been used to safeguard environments in which access control is of paramount importance. Security checkpoints have been placed at the entrances to many public places to detect prohibited items such as handguns and explosives. Human operators complete these tasks because automated recognition in baggage inspection is far from perfect. Research and development on X-ray testing is, however, ongoing into new approaches that can be used to aid human operators. This paper attempts to make a contribution to the field of object recognition by proposing a new approach called Adaptive Sparse Representation (XASR+). It consists of two stages: learning and testing. In the learning stage, for each object of training dataset, several patches are extracted from its X-ray images in order to construct representative dictionaries. A stop-list is used to remove very common words of the dictionaries. In the testing stage, test patches of the test image are extracted, and for each test patch a dictionary is built concatenating the 'best' representative dictionary of each object. Using this adapted dictionary, each test patch is classified following the Sparse Representation Classification (SRC) methodology. Finally, the test image is classified by patch voting. Thus, our approach is able to deal with less constrained conditions including some contrast variability, pose, intra-class variability, size of the image and focal distance. We tested the effectiveness of our method for the detection of four different objects. In our experiments, the recognition rate was more than 97% in each class, and more than 94% if the object is occluded less than 15%. Results show that XASR+ deals well with unconstrained conditions, outperforming various representative methods in the literature. Index Terms-X-ray testing, computer vision, sparse representations, image analysis.
Although unconstrained face recognition has been widely studied over recent years, state-of-the-a... more Although unconstrained face recognition has been widely studied over recent years, state-of-the-art algorithms still result in an unsatisfactory performance for low-quality images. In this paper, we make two contributions to this field: the first one is the release of a new dataset called 'AR-LQ' that can be used in conjunction with the well-known 'AR' dataset to evaluate face recognition algorithms on blurred and low-resolution face images. The proposed dataset contains five new blurred faces (at five different levels, from low to severe blurriness) and five new low-resolution images (at five different levels, from 66 ⇥ 48 to 7 ⇥ 5 pixels) for each of the hundred subjects of the 'AR' dataset. The new blurred images were acquired by using a DLSR camera with manual focus that takes an out-of-focus photograph of a monitor that displays a sharp face image. In the same way, the low-resolution images were acquired from the monitor by a DLSR at different distances. Thus, an attempt is made to acquire low-quality images that have been degraded by a real degradation process. Our second contribution is an extension of a known face recognition technique based on sparse representations (ASR) that takes into account low-resolution face images. The proposed method, called blur-ASR or bASR, was designed to recognize faces using dictionaries with different levels of blurriness. These were obtained by digitally blurring the training images, and a sharpness metric for matching blurriness between the query image and the dictionaries. These two main adjustments made the algorithm more robust with respect to low-quality images. In our experiments, bASR consistently outperforms other state-of-art methods including hand-crafted features, sparse representations, and a seven well-known deep learning face recognition techniques with and without super resolution techniques. On average, bASR obtained 88.8% of accuracy, whereas the rest obtained less than 78.4%.
Although face recognition systems have achieved impressive performance in recent years, the low-r... more Although face recognition systems have achieved impressive performance in recent years, the low-resolution face recognition task remains challenging, especially when the low-resolution faces are captured under non-ideal conditions, as is common in surveillance-based applications. Faces captured in such conditions are often contaminated by blur, non-uniform lighting, and non-frontal face pose. In this paper, we analyze face recognition techniques using data captured under low-quality conditions in the wild. We provide a comprehensive analysis of experimental results for two of the most important applications in real surveillance applications, and demonstrate practical approaches to handle both cases that show promising performance. The following three contributions are made: (i) we conduct experiments to evaluate super-resolution methods for low-resolution face recognition; (ii) we study face re-identification on various public face datasets including real surveillance and low-resolution subsets of large-scale datasets, present a baseline result for several deep learning based approaches, and improve them by introducing a Generative Adversarial Network (GAN) pre-training approach and fully convolutional architecture; and (iii) we explore low-resolution face identification by employing a state-of-the-art supervised discriminative learning approach. Evaluations are conducted on challenging portions of the SCface and UCCSface datasets.
X-ray testing of complex objects -such as luggage screening at airports-is usually performed manu... more X-ray testing of complex objects -such as luggage screening at airports-is usually performed manually. This is not always effective, since it depends strongly on the pose of the objects of interest (target objects), occlusion and human capabilities as well. Additionally, certain target objects are difficult to be detected using only one viewpoint. For this reason, we developed an active X-ray testing framework that is able to adequate the viewpoint of the target object in order to obtain better X-ray images to analyze. The key idea of our method is to adapt automatically the viewpoint of the X-ray images in order to project the target object in poses where the detection performance should be higher. Thus, the detection inside of complex objects can be performed in a more effective way. Using a robotic arm and a semi-automatic manipulator system, the robustness and reliability of the method have been verified in the automated detection of razor blades located inside of nine different objects showing promising preliminary results: in 130 experiments we were able to detect 115 times the razor blade with 10 false alarms, achieving recall of 89% and precision of 92%.
In countries where fish is often consumed, fish bones are some of the most frequently ingested fo... more In countries where fish is often consumed, fish bones are some of the most frequently ingested foreign bodies encountered in foods. In the production of fish fillets, fish bone detection is performed by human inspection using their sense of touch and vision which can lead to misclassification. Effective detection of fish bones in the quality control process would help avoid this problem. For this reason, an X-ray machine vision approach to automatically detect fish bones in fish fillets was developed. This paper describes our approach and the corresponding experiments with salmon and trout fillets. In the experiments, salmon X-ray images using 10×10 pixels detection windows and 24 intensity features (selected from 279 features) were analyzed. The methodology was validated using representative fish bones and trouts provided by a salmon industry and yielded a detection performance of 99%. We believe that the proposed approach opens new possibilities in the field of automated visual inspection of salmon, trout and other similar fish.
This paper describes the main activities and achievements of our research group on Machine Intell... more This paper describes the main activities and achievements of our research group on Machine Intelligence and Robotics (Grima) at the Computer Science Department, Pontificia Universidad Catolica de Chile (PUC). Since 2002, we have been developing an active research in the area of indoor autonomous social robots. Our main focus has been the cognitive side of Robotics, where we have developed algorithms for autonomous navigation using wheeled robots, scene recognition using vision and 3D range sensors, and social behaviors using Markov Decision Processes, among others. As a distinguishing feature, in our research we have followed a probabilistic approach, deeply rooted in machine learning and Bayesian statistical techniques. Among our main achievements are an increasing list of publications in main Robotics conference and journals, and the consolidation of a research group with more than 25 people among full-time professors, visiting researchers, and graduate students.
The classic image processing method for flaw detection uses one image of the scene, or multiple i... more The classic image processing method for flaw detection uses one image of the scene, or multiple images without correspondences between them. To improve this scheme, automated inspection using multiple views has been developed in recent years. This strategy's key idea is to consider as real flaws those regions that can be tracked in a sequence of multiple images because they are located in positions dictated by geometric conditions. In contrast, false alarms (or noise) can be successfully eliminated in this manner, since they do not appear in the predicted places in the following images, and thus cannot be tracked. This paper presents a method to inspect aluminum wheels using images taken from different positions by using a method called automatic multiple view inspection (AMVI). Our method can be applied to uncalibrated image sequences, therefore it is not necessary to determine optical and geometric parameters normally present in the calibrated systems. In addition, to improve the performance, we designed a false alarm reduction method in two and three views called Intermediate Classifier Block (ICB). The ICB method takes advantage of the classifier ensemble methodology by making use of feature analysis in multiple views. Using this method, real flaws can be detected with high precision while most false alarms can be discriminated.
Substantial research has been performed on automated detection and classification of welding defe... more Substantial research has been performed on automated detection and classification of welding defects in continuous welds using X-ray imaging. Typically, the detection follows a pattern recognition schema (segmentation, feature extraction and classification). In computer vision community, however, many object detection and classification problems, like face and human detection, have been recently solvedwithout segmentation-using sliding-windows and novel features like local binary patterns extracted from saliency maps. For this reason, we propose in this paper the use of sliding-windows with the mentioned features to perform automatically the automated detection of welding defects. In the experiments, we analyzed 5000 detection windows (24x24 pixels) and 572 intensity features from 10 representative X-ray images. Cross validation yielded a detection performance of 94% using a support vector machine classifier with only 14 selected features. The method was implemented and tested on real X-ray images showing high effectiveness. We believe that the proposed approach opens new possibilities in the field of automated detection of welding defects.
Deep-fat frying is a unit operation which develops unique sensorial attributes in foods. For inst... more Deep-fat frying is a unit operation which develops unique sensorial attributes in foods. For instance, texture is the principal quality parameter of tortilla and corn chips. On the other hand, computer vision is a useful tool for quality evaluation and prediction of some physical properties in different either raw or processed foods. The objective of this research was to characterize corn and tortilla chips by using computer vision, and to build proper mathematical models which permit to predict mechanical properties of these chips (maximum force, such as hardness, and distance to maximum force, such as toughness) by using chromatic features extracted from their corresponding digital images. Corn and tortilla chips (thickness of 2 mm; diameter of 37 mm) were made from masa of maize and fried at constant oil temperatures of 160, 175, and 190°C. A high linear correlation (R 2 >0.9400) was obtained between mechanical properties and some image features (Hu, Fourier, and Haralick moments). Cross-validation technique demonstrated the repeatability and good performance (>90%) of the models tested, indicating that can be used to predict the textural properties of the tortilla and corn chips by using selected features extracted from their digital images, without the necessity of measuring them in a texture analyzer.
X-ray testing of complex objects -such as luggage screening at airports-is usually performed manu... more X-ray testing of complex objects -such as luggage screening at airports-is usually performed manually. This is not always effective, since it depends strongly on the pose of the objects of interest (target objects), occlusion and human capabilities as well. Additionally, certain target objects are difficult to be detected using only one viewpoint. For this reason, we developed an active X-ray testing framework that is able to adequate the viewpoint of the target object in order to obtain better X-ray images to analyze. The key idea of our method is to adapt automatically the viewpoint of the X-ray images in order to project the target object in poses where the detection performance should be higher. Thus, the detection inside of complex objects can be performed in a more effective way. Using a robotic arm and a semi-automatic manipulator system, the robustness and reliability of the method have been verified in the automated detection of razor blades located inside of nine different objects showing promising preliminary results: in 130 experiments we were able to detect 115 times the razor blade with 10 false alarms, achieving recall of 89% and precision of 92%.
Considerable research efforts in computer vision applied to food quality evaluation have been dev... more Considerable research efforts in computer vision applied to food quality evaluation have been developed in the last years; however, they have been concentrated on using or developing tailored methods based on visual features that are able to solve a specific task. Nevertheless, today's computer capabilities are giving us new ways to solve complex computer vision problems. In particular, a new paradigm on machine learning techniques has emerged posing the task of recognizing visual patterns as a search problem based on training data and a hypothesis space composed by visual features and suitable classifiers. Furthermore, now we are able to extract, process, and test in the same time more image features and classifiers than before. Thus, we propose a general framework that designs a computer vision system automatically, i.e., it findswithout human interaction-the features and the clas-sifiers for a given application avoiding the classical trial and error framework commonly used by human designers. The key idea of the proposed framework is to select-automatically-from a large set of features and a bank of classifiers those features and classifiers that achieve the highest performance. We tested our framework on eight different food quality evaluation problems yielding a classification performance of 95 % or more in every case. The proposed framework was implemented as a Matlab Toolbox available for noncommercial purposes.
In order to reduce the security risk of a commercial aircraft, passengers are not allowed to take... more In order to reduce the security risk of a commercial aircraft, passengers are not allowed to take certain items in carry-on baggage. For this reason, human operators are trained to detect prohibited items using a manually controlled baggage screening process. In this paper, we propose the use of a method based on multiple X-ray views to detect some regular prohibited items with very defined shapes and sizes. The method consists of two steps: 'structure estimation', to obtain a geometric model of the multiple views from the object to be inspected (a baggage), and 'parts detection', to detect the parts of interest (prohibited items). The geometric model is estimated using a structure from motion algorithm. The detection of the parts of interest is performed by an adhoc segmentation algorithm (object dependent) followed by a general tracking algorithm based on geometric and appearance constraints. In order to illustrate the effectiveness of the proposed method, experimental results on detecting regular objects −razor blades and guns− are shown yielding promising results.
This paper presents a new methodology for identifying parts of interest inside of a complex objec... more This paper presents a new methodology for identifying parts of interest inside of a complex object using multiple X-ray views. The proposed method consists of five steps: A) image acquisition, that acquires an image sequence where the parts of the object are captured from different viewpoints; B) geometric model estimation, that establishes a multiple view geometric model used to find the correct correspondence among different views; C) single view detection, that segment potential regions of interest in each view; D) multiple view detection, that matches and tracks potential regions based on similarity and geometrical multiple view constraints; and E) analysis, that analyzes the tracked regions using multiple view information, filtering out false alarms without eliminating existing parts of interest. In order to evaluate the effectiveness of the proposed method, the algorithm was tested on 32 cases (five applications using different segmentation approaches) yielding promising results: precision and recall were 95.7% and 93.9%, respectively. Additionally, the multiple view information obtained from the tracked parts was effectively used for recognition purposes. In our recognition experiments, we obtained an accuracy of 96.5%. Validation experiments show that our approach achieves better performance than other representative methods in the literature.
X-ray testing has been developed for inspection of materials or objects, where the aim is to anal... more X-ray testing has been developed for inspection of materials or objects, where the aim is to analyze -nondestructively-those inner parts that are undetectable to the naked eye. Thus, Xray testing is used to determine if a test object deviates from a given set of specifications. Typical applications are inspection of automotive parts, quality control of welds, screening of baggage, analysis of food products and inspection of cargos. In order to achieve efficient and effective X-ray testing, automated and semi-automated systems based on computer vision algorithms are being developed to execute this task. In this paper, we present a general overview of computer vision approaches that have been used in X-ray testing. In addition, we review some techniques that have been applied in certain relevant applications; and we introduce a public database of X-ray images that can be used for testing and evaluation of image analysis and computer vision algorithms. Finally, we conclude that there are some areas -like casting inspection where automated systems are very effective, and other application areas -such as baggage screening-where human inspection is still used; there are certain application areas -like weld and cargo inspections-where the process is semiautomatic; and there is some research in areas -including food analysis-where processes are beginning to be characterized by the use of X-ray imaging.
In order to reduce the security risk of a commercial aircraft, passengers are not allowed to take... more In order to reduce the security risk of a commercial aircraft, passengers are not allowed to take certain items in their carry-on baggage. For this reason, human operators are trained to detect prohibited items using a manually controlled baggage screening process. The inspection process, however, is highly complex as hazardous items are very difficult to detect when placed in close packed bags, superimposed by other objects, and/or rotated showing an unrecognizable profile. In this paper, we review certain advances achieved by our research group in this field. Our methodology is based on multiple view analysis, because it can be a powerful tool for examining complex objects in cases in which uncertainty can lead to misinterpretation. In our approach, multiple views (taken from fixed points of view, or using an active vision approach in which the best views are automated selected) are analyzed in the detection of regular objects. In order to illustrate the effectiveness of the proposed method, experimental results on recognizing guns, razor blades, pins, clips and springs in baggage inspection are presented achieving around 90% accuracy. We believe that it would be possible to design an automated aid in a target detection task using the proposed algorithm.
Baggage inspection using X-ray screening is a priority task that reduces the risk of crime and te... more Baggage inspection using X-ray screening is a priority task that reduces the risk of crime and terrorist attacks. Manual detection of threat items is tedious because very few bags actually contain threat items and the process requires a high degree of concentration. An automated solution would be a welcome development in this field. We propose a methodology for automatic detection of threat objects using single X-ray images. Our approach is an adaptation of a methodology originally created for recognizing objects in photographs based on Implicit Shape Models. Our detection method uses a visual vocabulary and an occurrence structure generated from a training dataset that contains representative X-ray images of the threat object to be detected. Our method can be applied to single views of grayscale X-ray images obtained using a single energy acquisition system. We tested the effectiveness of our method for the detection of three different threat objects: razor blades, shuriken (ninja stars) and handguns. The testing dataset for each threat object consisted of 200 X-ray images of bags. The true positive and false positive rates (TPR, FPR) are: (0.99, 0.02) for razor blades, (0.97, 0.06) for shuriken and (0.89, 0.17) for handguns. If other representative training datasets were utilized, we believe that our methodology could aid in the detection of other kinds of threat objects.
In this paper, we present a new dataset consisting of 19,407 X-ray images. The images are organiz... more In this paper, we present a new dataset consisting of 19,407 X-ray images. The images are organized in a public database called GDXray that can be used free of charge, but for research and educational purposes only. The database includes five groups of X-ray images: castings, welds, baggage, natural objects and settings. Each group has several series, and each series several X-ray images. Most of the series are annotated or labeled. In such cases, the coordinates of the bounding boxes of the objects of interest or the labels of the images are available in standard text files. The size of GDXray is 3.5 GB and it can be downloaded from our website. We believe that GDXray represents a relevant contribution to the X-ray testing community. On the one hand, students, researchers and engineers can use these X-ray images to develop, test and evaluate image analysis and computer vision algorithms without purchasing expensive X-ray equipment. On the other hand, these images can be used as a benchmark in order to test and compare the performance of different approaches on the same data. Moreover, the database can be used in the training programs of human inspectors.
It is well known that some facial attributes -like soft biometric traits-can increase the perform... more It is well known that some facial attributes -like soft biometric traits-can increase the performance of traditional biometric systems and help recognition based on human descriptions. In addition, other facial attributes, such as facial expressions, can be used in human-computer interfaces, image retrieval, talking heads and human emotion analysis. This paper addresses the problem of automated recognition of facial attributes by proposing a new general approach called Adaptive Sparse Representation of Random Patches (ASR+). The proposed method consists of two stages: In the learning stage, random patches are extracted from representative face images of each class (e.g., in gender recognition -a two-class problem-, images of females/males) in order to construct representative dictionaries. A stop list is used to remove very common words of the dictionaries. In the testing stage, random test patches of the query image are extracted, and for each non-stopped test patch a dictionary is built concatenating the 'best' representative dictionary of each class. Using this adapted dictionary, each non-stopped test patch is classified following the Sparse Representation Classification (SRC) methodology. Finally, the query image is classified by patch voting. Thus, our approach is able to learn a model for each recognition task dealing with a larger degree of variability in ambient lighting, pose, expression, occlusion, face size and distance from the camera. Experiments were carried out on eight face databases in order to recognize facial expression, gender, race, disguise and beard. Results show that ASR+ deals well with unconstrained conditions, outperforming various representative methods in the literature in many complex scenarios.
Minimum Norm Estimation (MNE) is an inverse solution method widely used to reconstruct the source... more Minimum Norm Estimation (MNE) is an inverse solution method widely used to reconstruct the source time series that underlie magnetoencephalography (MEG) data. MNE addresses the ill-posed nature of MEG source estimation through regularization (e.g., Tikhonov regularization). Selecting the best regularization parameter is a critical step. Generally, once set, it is common practice to keep the same coefficient throughout a study. However, it is yet to be known whether the optimal lambda for spectral power analysis of MEG source data coincides with the optimal regularization for source-level oscillatory coupling analysis. We addressed this question via extensive Monte-Carlo simulations of MEG data, where we generated 21,600 configurations of pairs of coupled sources with varying sizes, signal-to-noise ratio (SNR), and coupling strengths. Then, we searched for the Tikhonov regularization coefficients (lambda) that maximize detection performance for (a) power and (b) coherence. For coherence, the optimal lambda was two orders of magnitude smaller than the best lambda for power. Moreover, we found that the spatial extent of the interacting sources and SNR, but not the extent of coupling, were the main parameters affecting the best choice for lambda. Our findings suggest using less regularization when measuring oscillatory coupling compared to power estimation.
In recent years, X-ray screening systems have been used to safeguard environments in which access... more In recent years, X-ray screening systems have been used to safeguard environments in which access control is of paramount importance. Security checkpoints have been placed at the entrances to many public places to detect prohibited items such as handguns and explosives. Human operators complete these tasks because automated recognition in baggage inspection is far from perfect. Research and development on X-ray testing is, however, ongoing into new approaches that can be used to aid human operators. This paper attempts to make a contribution to the field of object recognition by proposing a new approach called Adaptive Sparse Representation (XASR+). It consists of two stages: learning and testing. In the learning stage, for each object of training dataset, several patches are extracted from its X-ray images in order to construct representative dictionaries. A stop-list is used to remove very common words of the dictionaries. In the testing stage, test patches of the test image are extracted, and for each test patch a dictionary is built concatenating the 'best' representative dictionary of each object. Using this adapted dictionary, each test patch is classified following the Sparse Representation Classification (SRC) methodology. Finally, the test image is classified by patch voting. Thus, our approach is able to deal with less constrained conditions including some contrast variability, pose, intra-class variability, size of the image and focal distance. We tested the effectiveness of our method for the detection of four different objects. In our experiments, the recognition rate was more than 97% in each class, and more than 94% if the object is occluded less than 15%. Results show that XASR+ deals well with unconstrained conditions, outperforming various representative methods in the literature. Index Terms-X-ray testing, computer vision, sparse representations, image analysis.
Although unconstrained face recognition has been widely studied over recent years, state-of-the-a... more Although unconstrained face recognition has been widely studied over recent years, state-of-the-art algorithms still result in an unsatisfactory performance for low-quality images. In this paper, we make two contributions to this field: the first one is the release of a new dataset called 'AR-LQ' that can be used in conjunction with the well-known 'AR' dataset to evaluate face recognition algorithms on blurred and low-resolution face images. The proposed dataset contains five new blurred faces (at five different levels, from low to severe blurriness) and five new low-resolution images (at five different levels, from 66 ⇥ 48 to 7 ⇥ 5 pixels) for each of the hundred subjects of the 'AR' dataset. The new blurred images were acquired by using a DLSR camera with manual focus that takes an out-of-focus photograph of a monitor that displays a sharp face image. In the same way, the low-resolution images were acquired from the monitor by a DLSR at different distances. Thus, an attempt is made to acquire low-quality images that have been degraded by a real degradation process. Our second contribution is an extension of a known face recognition technique based on sparse representations (ASR) that takes into account low-resolution face images. The proposed method, called blur-ASR or bASR, was designed to recognize faces using dictionaries with different levels of blurriness. These were obtained by digitally blurring the training images, and a sharpness metric for matching blurriness between the query image and the dictionaries. These two main adjustments made the algorithm more robust with respect to low-quality images. In our experiments, bASR consistently outperforms other state-of-art methods including hand-crafted features, sparse representations, and a seven well-known deep learning face recognition techniques with and without super resolution techniques. On average, bASR obtained 88.8% of accuracy, whereas the rest obtained less than 78.4%.
Although face recognition systems have achieved impressive performance in recent years, the low-r... more Although face recognition systems have achieved impressive performance in recent years, the low-resolution face recognition task remains challenging, especially when the low-resolution faces are captured under non-ideal conditions, as is common in surveillance-based applications. Faces captured in such conditions are often contaminated by blur, non-uniform lighting, and non-frontal face pose. In this paper, we analyze face recognition techniques using data captured under low-quality conditions in the wild. We provide a comprehensive analysis of experimental results for two of the most important applications in real surveillance applications, and demonstrate practical approaches to handle both cases that show promising performance. The following three contributions are made: (i) we conduct experiments to evaluate super-resolution methods for low-resolution face recognition; (ii) we study face re-identification on various public face datasets including real surveillance and low-resolution subsets of large-scale datasets, present a baseline result for several deep learning based approaches, and improve them by introducing a Generative Adversarial Network (GAN) pre-training approach and fully convolutional architecture; and (iii) we explore low-resolution face identification by employing a state-of-the-art supervised discriminative learning approach. Evaluations are conducted on challenging portions of the SCface and UCCSface datasets.
Uploads
Papers by Domingo Mery