A fully automatic redeye detection and correction algorithm

A fully automatic redeye detection and correction algorithm is presented to address the redeye artifacts in digital photos. The algorithm contains a redeye detection part and a correction part. The detection part is modeled as a feature based object detection problem. Adaboost is used to simultaneously select features and train the classifier. A new feature set is designed to address the orientation-dependency problem associated with the Haar-like features commonly used for object detection design. For each detected redeye, a correction algorithm is applied to do adaptive desaturation and darkening over the redeye region.

An algorithm for the daily localization of the prostate using implanted markers and a standard video-based electronic portal imaging device (V-EPID) has been tested. Prior to planning, three gold markers were implanted in the prostate of seven patients. The clinical images were acquired with a BeamViewPlus 2.1 V-EPID for each field during the normal course radiotherapy treatment and are used off-line to determine the ability of the automatic marker detection algorithm to adequately and consistently detect the markers. Clinical images were obtained with various dose levels from ranging 2.5 to 75 MU. The algorithm is based on marker attenuation characterization in the portal image and spatial distribution. A total of 1182 clinical images were taken. The results show an average efficiency of 93% for the markers detected individually and 85% for the group of markers. This algorithm accomplishes the detection and validation in 0.20-0.40 s. When the center of mass of the group of implanted markers is used, then all displacements can be corrected to within 1.0 mm in 84% of the cases and within 1.5 mm in 97% of cases. The standard video-based EPID tested provides excellent marker detection capability even with low dose levels. The V-EPID can be used successfully with radiopaque markers and the automatic detection algorithm to track and correct the daily setup deviations due to organ motions.

Owing to advances in information technology, some studies have been conducted on work environment monitoring system using various sensors in order to prevention of accidents, information gathering, and optimal management of work environment. It was found that a factory environment monitoring system using climate sensors such as O2, Co2, Nh3, and Pm2 has recently become essential because it can help protect the safety and health of workers. Climate sensors are a mesh-type arrangement placed at certain distances apart to acquire and anlyze exact environmental information. Sensitivity, specificity, and accuracy increase as the number of sensors are increased. However, as the number of sensors increases, it becomes more difficult to detect faulty sensors, and in the worst case, false information can lead to accidents. It is necessary, therefore, for environment monitoring systems using a large number of climate sensors to have a function that will automatically detect the failure of a sensor. The value of an individual climate sensor is organically realted to the value of neighbor sensors, unless they are located in enclosed spaces. If the sensor value at a specific position changes, the neighboring sensor's values are also changed. In the past, much research has studied algorithms to improve the sensing accuracy of specific location using neighbor sensor data based on these principle. If these algorithms are used inversely, it is possible to infer or predict the environmental information in the area where the sensor has failed, by using the values from neighboring sensors. Even with these systems in operaton, a major concern on many industrial sites is that work does not stop even if a sensor failure is detected. In other words, when workers are in areas where a sensor has failed, they may become exposed to hazardous conditions. Therefore, even if a sensor fails, for the sake of the workers, it is necessary to continuously provide environmental information to those in the affected area. This paper presents the automatic sensor fault detection and sensor reconstruction algorithm for emergency recovery relative to the production of continuous and reliable environmental data. The principle of automatic sensor fault detection and the sensor reconstruction algorithm are the same. The proposed algorithms consist of four steps. In the first step, a total of nine sensors consisting of 3∗3 are configured as one set. In the second step, the three sensors, including the central sensor, are grouped into one group. One set becomes a total of four groups. In the third step, reference curve maps (RCM) are created to record changes in sensor values according to the amount of ambient gas. The RCM records the sensor's values as the gas changes in density. Four RCMs are created per set. A total of 32 RCMs are created because one sensor is included in a total of eight sets. In the fourth step, the automatic sensor fault detection and sensor reconstruction algorithms are performed. Automatic sensor fault detection consists of two substeps. In the first sub-step, the comparative values of the failed sensor and the neighboring sensor are added to the RCM. In the second substep, if more than half of the comparison result deviates from the normal range, the target sensor is judged to be faulty, and the supervisor is notified. If the superviisor determines that the sensor is normal, the RCMs are then updated to improve accuracy. Sensor reconstruction is performed when the sensor is determined to be faulty. Sensor reconstruction consists of two further substeps. First, the failure of the sensor is inferred using all RCMs and the linear interpolation technique. Second, the final value of the failed sensor is determined by using its predicted value, as obtained by using RCMs and weighted averages.

The paper proposes a technology for automated video-based observation (VBO) of a drogue-sensor in the problem of aerial refueling. The technology is based on the use of a passive optoelectronic system and incorporates the logic of automated refueling observation of a refueling process using algorithms for the automatic detection and tracking of a drogue-sensor, a methodical apparatus for suboptimal linear filtering of the observed process under the conditions of spatial and temporary non-stationarity of the refueling process, algorithms for automatic correlation detection and tracking of a drogue-sensor using suboptimal filtering. An analysis of the design of experimental foreign systems for autonomous aerial refueling is carried out. The choice of the algorithm for the functioning of the synthetic vision system is substantiated. It is established that the main observation procedures: detection, capture for tracking and determination of the current drogue coordinates with a given rate and quality should be performed automatically, the pilot-operator takes part in the operation of the synthetic vision system in case of capture errors or mistracking. The statement of the problem for automated VBO of a drogue-sensor is formulated. A structural-logical diagram of the automated observation process, including the detection and tracking of a drogue, as well as decision-making by the pilot in various situations, is proposed. A modeling complex for a synthetic vision system operation is presented. The results of experimental studies of the synthetic vision system efficiency are presented. Based on the developed technology and the results of evaluating the effectiveness of automated observation algorithms, a strategy for performing autonomous refueling in conditions of various turbulence is proposed, while, during weak turbulence, a successful engagement is provided by tracking the center of drogue oscillations, in turn, under conditions of severe turbulence, a successful engagement can be provided by tracking a drogue controlled according to the synthetic vision system data.

The detection of (moving or static) targets in remotely sensed hyperspectral images often requires real-time responses for swift decisions that depend upon high computing performance of algorithm analysis. The automatic target detection and classification algorithm (ATDCA) has been widely used for this purpose. In this letter, we develop several optimizations for accelerating the computational performance of ATDCA. The first one focuses on the use of the Gram-Schmidt orthogonalization method instead of the orthogonal projection process adopted by the classic algorithm. The second one is focused on the development of a new implementation of the algorithm on commodity graphics processing units (GPUs). The proposed GPU implementation properly exploits the GPU architecture at low level, including shared memory, and provides coalesced accesses to memory that lead to very significant speedup factors, thus taking full advantage of the computational power of GPUs. The GPU implementation is specifically tailored to hyperspectral imagery and the special characteristics of this kind of data, achieving real-time performance of ATDCA for the first time in the literature. The proposed optimizations are evaluated not only in terms of target detection accuracy but also in terms of computational performance using two different GPU architectures by NVIDIA: Tesla C1060 and GeForce GTX 580, taking advantage of the performance of operations in single-precision floating point. Experiments are conducted using hyperspectral data sets collected by three different hyperspectral imaging instruments. These results reveal considerable acceleration factors while retaining the same target detection accuracy for the algorithm.

An automatic target detection algorithm is developed for infrared image small target in complicate background of sea and sky. Wavelet multiresolution edge detection algorithm is adopted to detect the sea-level line from coarse to fine. A strip can be decided accordingly as the potential area where the naval vessel targets in infrared images usually appear. We realized target detection by defining an energy function that integrated the results of horizontal wavelet transform. We complete the detection by marking the position where the energy is assumed the maximum and improving the target location precision by comparing the dissimilarity between a couple of windows. Experiment results indicate that the method can detect and locate small targets precisely in an infrared image with high detection probability.

An automatic detection algorithm for dead birds based on support vector machine (SVM) is proposed. Firstly, according to the changes of central region of cockscomb in the picture, logic and operation is used to remove the image of live chickens; Secondly, in order to distinguish accurately the dead birds in processed picture, the perimeter, area, eccentricity and complexity of the cockscomb are extracted as the variables. The changes of these variables are defined as the feature vectors. The samples of the above feature vectors are used to train SVM classifier. During the training, the grid search method is used to optimize the kernel width and punishment factor of SVM and the classifier for dead bird is designed finally. The results of experiment show that the detection accuracy is over 90%.

Alternans, a beat-to-beat alternation in cardiac signals, may serve as a precursor to lethal cardiac arrhythmias, including ventricular tachycardia and ventricular fibrillation. Therefore, alternans is a desirable target of early arrhythmia prediction/detection. For long-term records and in the presence of noise, the definition of alternans is qualitative and ambiguous. This makes their automatic detection in large spatiotemporal data sets almost impossible. We present here a quantitative combinatorics-derived definition of alternans in the presence of random noise and a novel algorithm for automatic alternans detection using criteria like temporal persistence (TP), representative phase (RP) and alternans ratio (AR). This technique is validated by comparison to theoretically-derived probabilities and by test data sets with white noise. Finally, the algorithm is applied to ultra-high resolution optical mapping data from cultured cell monolayers, exhibiting calcium alternans. Early fine-scale alternans, close to the noise level, were revealed and linked to the later formation of larger regions and evolution of spatially discordant alternans (SDA). This robust new technique can be useful in quantification and better understanding of the onset of arrhythmias and in general analysis of space-time alternating signals.

The diseases and injuries of the knee joint are the most common orthopedic disorders. Personalized knee models can be helpful in the process of early intervention and lasting treatment techniques development. Fully automatic reconstruction of knee joint anatomical structures from medical images (CT, MRI, ultrasound) remains a challenge. For this reason, most of state-of-the-art knee joint models contain simplifications such as representation of muscles and ligaments as line segments connecting two points which replace attachment areas. The paper presents algorithms for automatic detection of such points on knee CT images. This paper presents three approaches to automatic detection of ligaments and tendons attachment sites on the patients CT images: qualitative anatomical descriptions, analysis of bones curvature, and quantitative anatomical descriptions. Combinations of these approaches result in new automatic detection algorithms. Each algorithm exploits anatomical peculiarities of each attachment site, e.g., bone curvature and number of other attachments in a neighborhood of the site. The experimental dataset consisted of 26 anonymized CT sequences containing right and left knee joints in different resolutions. The proposed algorithms take into account bone surface curvatures and spatial differences in locations of medial and lateral parts of both knees. The algorithms for detection of quadriceps femoris, popliteus, biceps femoris tendons, and lateral collateral and medial collateral ligaments attachment sites are provided, as well as examples of their application. Two algorithms are validated by comparison with known statistics of ligaments lengths and also using ground truth annotations for anatomical landmarks approved by clinical experts. The algorithms simplify generation of patient-specific knee joint models demanded in personalized biomechanical models. The algorithms in the current implementation have two important limitations. First, the correctness of the produced results depends on the bones segmentation quality. Second, the presented algorithms detect a point of the attachment site, which is not necessarily its center. Therefore, manual correction of the attachment site location may be required for attachments with relatively large area.

Considering the fact that the amount of traffic incident data is rare compared to the large amount of normal traffic state data in the real word, we proposed an Automatic Incident Detection (AID) algorithm based on subsampling method. First, an improved subsampling method based on Edited Nearest Neighbor Rule (ENN) algorithm was used to reconstruct the training set to get a balanced dataset. Then, the Support Vector Machine (SVM) was adopted as a classifier to detect traffic incidents. The real traffic data collected from the I-880 freeway in American was used to build the model and test the performance of the proposed AID algorithm. In addition, we made a comparison of the detection performances between the AID algorithm obtained by the original training set and the one by the relatively balanced training set. The experimental results show that the proposed AID algorithm based on subsampling is suitable for imbalanced dataset and can obtain a better detection performance.

Automatic detection and recognition of targets by means of passive IR sensors suffer from limitations due to lack of sufficient contrast between the targets and their background, and among the facets of a target. In this paper the results of a suite of polarization-sensitive automatic target detection and recognition algorithms on sets of simulated and real polarimetric IR imagery are presented. A custom designed Polarimetric IR (PIR) imaging sensor is used for collecting real polarimetric target data-three of the four Stokes parameters under a variety of conditions. Then a set of novel algorithms are designed and tested that uses the target and background Stokes parameters for detection, segmentation and classification of targets. The empirical performance results are obtained in terms of the probabilities of detection, false alarm rate, segmentation accuracy, and recognition probabilities as functions of number of pixels on target, aspect and depression angles and under several background conditions (clutter densities) on the polarimetric and non-polarimetrirc data. These results show that a noticeable improvement over the non-polarimetric ATR can be achieved.

Asthma is a symptom of tracheal obstruction caused by bronchospasm, and it is among the most prevalent chronic obstructive pulmonary diseases. Auscultation is the most commonly used approach for the clinical diagnosis of asthma. However, recognizing wheezes through auscultation requires experienced physicians, and this approach is not sufficiently objective. Therefore, developing a method for recognizing wheezes objectively is crucial. Most studies have used the spectral features of lung sounds to detect wheezes; however, they have not achieved sufficiently high performance owing to the poor discrimination of spectral features. Several studies have attempted to extract wheezing features from lung sound spectrograms; however, their approaches were easily affected by variations in the wheezing frequency and background noise. The present study proposes a novel automatic wheeze detection algorithm for extracting lung sound features in the time–frequency domain and automatically detecting wheezes. The proposed algorithm applies canonical correlation analysis to successfully detect wheezing features in a lung sound spectrogram. Moreover, a neural network technique is used to effectively classify healthy and wheezing sounds. The experimental results indicated that the proposed algorithm showed excellent performance in detecting wheezing.

Aiming at the common grammatical errors in English writing, this paper constructs a large-scale corpus containing various grammatical errors, and cleans and classifies the data through manual annotation. Then, a deep learning model combining CNN (convolutional neural network) and RNN (recurrent neural network) is used to train the error types in the corpus to automatically identify and mark grammatical errors. Finally, the detection results output by the model are compared with the manual proofreading results to calculate the correction effect. The model has a high accuracy and recall rate in the detection of word form errors in long sentences, with an F1 value of 0.93. This study shows that the automatic detection and correction algorithm of grammatical errors based on deep learning has significant advantages in improving the writing accuracy of English learners.

The analysis of the impact crater size-frequency distribution (CSFD) is a well-established approach to the determination of the age of planetary surfaces. Classically, estimation of the CSFD is achieved by manual crater counting and size determination in spacecraft images, which, however, becomes very time-consuming for large surface areas and/or high image resolution. With increasing availability of high-resolution (nearly) global image mosaics of planetary surfaces, a variety of automated methods for the detection of craters based on image data and/or topographic data have been developed. In this contribution a template-based crater detection algorithm is used which analyses image data acquired under known illumination conditions. Its results are used to establish the CSFD for the examined area, which is then used to estimate the absolute model age of the surface. The detection threshold of the automatic crater detection algorithm is calibrated based on a region with available manually determined CSFD such that the age inferred from the manual crater counts corresponds to the age inferred from the automatic crater detection results. With this detection threshold, the automatic crater detection algorithm can be applied to a much larger surface region around the calibration area. The proposed age estimation method is demonstrated for a Kaguya Terrain Camera image mosaic of 7.4 m per pixel resolution of the floor region of the lunar crater Tsiolkovsky, which consists of dark and flat mare basalt and has an area of nearly 10,000 km<sup>2</sup>. The region used for calibration, for which manual crater counts are available, has an area of 100 km<sup>2</sup>. In order to obtain a spatially resolved age map, CSFDs and surface ages are computed for overlapping quadratic regions of about 4.4 x 4.4 km<sup>2</sup> size offset by a step width of 74 m. Our constructed surface age map of the floor of Tsiolkovsky shows age values of typically 3.2-3.3 Ga, while for small regions lower (down to 2.9 Ga) and higher (up to 3.6 Ga) age values can be observed. It is known that CSFD-derived absolute model ages can exhibit variations although the surface has a constant age. However, for four 10-20 km sized regions in the eastern part of the crater floor our map shows age values differing by several hundred Ma from the typical age of the crater floor, where the same regions are also discernible in Clementine UV/VIS color ratio image data probably due to compositional variations, such that the age differences of these four regions may be real.

Simple SummaryArtificial Intelligence (AI) algorithms can assist clinicians in their daily tasks by automatically detecting and/or classifying nodules in chest CT scans. Bias of such algorithms is one of the reasons why implementation of them in clinical practice is still not widely adopted. There is no published review on the bias that these algorithms may contain. This review aims to present different types of bias in such algorithms and present possible ways to mitigate them. Only then it would be possible to ensure that these algorithms work as intended under many different clinical settings.Artificial Intelligence (AI) algorithms for automatic lung nodule detection and classification can assist radiologists in their daily routine of chest CT evaluation. Even though many AI algorithms for these tasks have already been developed, their implementation in the clinical workflow is still largely lacking. Apart from the significant number of false-positive findings, one of the reasons for that is the bias that these algorithms may contain. In this review, different types of biases that may exist in chest CT AI nodule detection and classification algorithms are listed and discussed. Examples from the literature in which each type of bias occurs are presented, along with ways to mitigate these biases. Different types of biases can occur in chest CT AI algorithms for lung nodule detection and classification. Mitigation of them can be very difficult, if not impossible to achieve completely.