Predicting the Particle Size Distributions of Spherical Particle Sets from Synthetic Images: A Comparison of 9 Classic Image Features

The two objectives of this paper were to demonstrate use the of the discrete element method for generating synthetic images of spherical particle configurations, and to compare the performance of 9 classic feature extraction methods for predicting the particle size distributions (PSD) from these images. The discrete element code YADE was used to generate synthetic images of granular materials to build the dataset. Nine feature extraction methods were compared: Haralick features, Histograms of Oriented Gradients, Entropy, Local Binary Patterns, Local Configuration Pattern, Complete Local Binary Patterns, the Fast Fourier transform, Gabor filters, and Discrete Haar Wavelets. The feature extraction methods were used to generate the inputs of neural networks to predict the PSD. The results show that feature extraction methods can predict the percentage passing with a root-mean-square error (RMSE) on the percentage passing as low as 1.7%. CLBP showed the best result for all particle sizes with a RMSE of 3.8 %. Better RMSE were obtained for the finest sieve (2.1%) compared to coarsest sieve (5.2%).

Detection of Pitt–Hopkins Syndrome Based on Morphological Facial Features

This work describes a non-invasive, automated software framework to discriminate between individuals with a genetic disorder, Pitt–Hopkins syndrome (PTHS), and healthy individuals through the identification of morphological facial features. The input data consist of frontal facial photographs in which faces are located using histograms of oriented gradients feature descriptors. Pre-processing steps include color normalization and enhancement, scaling down, rotation, and cropping of pictures to produce a series of images of faces with consistent dimensions. Sixty-eight facial landmarks are automatically located on each face through a cascade of regression functions learnt via gradient boosting to estimate the shape from an initial approximation. The intensities of a sparse set of pixels indexed relative to this initial estimate are used to determine the landmarks. A set of carefully selected geometric features, for example, the relative width of the mouth or angle of the nose, is extracted from the landmarks. The features are used to investigate the statistical differences between the two populations of PTHS and healthy controls. The methodology was tested on 71 individuals with PTHS and 55 healthy controls. The software was able to classify individuals with an accuracy rate of 91%, while pediatricians achieved a recognition rate of 74%. Two geometric features related to the nose and mouth showed significant statistical difference between the two populations.

Color Based Object Categorization Using Histograms of Oriented Hue and Saturation

In the last few years, there has been a lot of interest in making smart components, e.g. robots, able to simulate human capacity of object recognition and categorization. In this paper, we propose a new revolutionary approach for object categorization based on combining the HOG (Histograms of Oriented Gradients) descriptors with our two new descriptors, HOH (Histograms of Oriented Hue) and HOS (Histograms of Oriented Saturation), designed it in the HSL (Hue, Saturation and Luminance) color space and inspired by this famous HOG descriptor. By using the chrominance components, we have succeeded in making the proposed descriptor invariant to all lighting conditions changes. Moreover, the use of this oriented gradient makes our descriptor invariant to geometric condition changes including geometric and photometric transformation. Finally, the combination of color and gradient information increase the recognition rate of this descriptor and give it an exceptional performance compared to existing methods in the recognition of colored handmade objects with uniform background (98.92% for Columbia Object Image Library and 99.16% for the Amsterdam Library of Object Images). For the classification task, we propose the use of two strong and very used classifiers, SVM (Support Vector Machine) and KNN (k-nearest neighbors) classifiers.

Improved Visual Recognition Memory Model Based on Grid Cells for Face Recognition

Traditional facial recognition methods depend on a large number of training samples due to the massive turning of synaptic weights for low-level feature extractions. In prior work, a brain-inspired model of visual recognition memory suggested that grid cells encode translation saccadic eye movement vectors between salient stimulus features. With a small training set for each recognition type, the relative positions among the selected features for each image were represented using grid and feature label cells in Hebbian learning. However, this model is suitable only for the recognition of familiar faces, objects, and scenes. The model's performance for a given face with unfamiliar facial expressions was unsatisfactory. In this study, an improved computational model via grid cells for facial recognition was proposed. Here, the initial hypothesis about stimulus identity was obtained using the histograms of oriented gradients (HOG) algorithm. The HOG descriptors effectively captured the sample edge or gradient structure features. Thus, most test samples were successfully recognized within three saccades. Moreover, the probability of a false hypothesis and the average fixations for successful recognition were reduced. Compared with other neural network models, such as convolutional neural networks and deep belief networks, the proposed method shows the best performance with only one training sample for each face. Moreover, it is robust against image occlusion and size variance or scaling. Our results may give insight for efficient recognition with small training samples based on neural networks.

Thyroid Nodule Classification in Ultrasound Images by Fusion of Conventional Features and Res-GAN Deep Features

In spite of the gargantuan number of patients affected by the thyroid nodule, the detection at an early stage is still a challenging task. Thyroid ultrasonography (US) is a noninvasive, inexpensive procedure widely used to detect and evaluate the thyroid nodules. The ultrasonography method for image classification is a computer-aided diagnostic technology based on image features. In this paper, we illustrate a method which involves the combination of the deep features with the conventional features together to form a hybrid feature space. Several image enhancement techniques, such as histogram equalization, Laplacian operator, logarithm transform, and Gamma correction, are undertaken to improve the quality and characteristics of the image before feature extraction. Among these methods, applying histogram equalization not only improves the brightness and contrast of the image but also achieves the highest classification accuracy at 69.8%. We extract features such as histograms of oriented gradients, local binary pattern, SIFT, and SURF and combine them with deep features of residual generative adversarial network. We compare the ResNet18, a residual convolutional neural network with 18 layers, with the Res-GAN, a residual generative adversarial network. The experimental result shows that Res-GAN outperforms the former model. Besides, we fuse SURF with deep features with a random forest model as a classifier, which achieves 95% accuracy.

Multi-Descriptor Random Sampling for Patch-Based Face Recognition

While there has been a massive increase in research into face recognition, it remains a challenging problem due to conditions present in real life. This paper focuses on the inherently present issue of partial occlusion distortions in real face recognition applications. We propose an approach to tackle this problem. First, face images are divided into multiple patches before local descriptors of Local Binary Patterns and Histograms of Oriented Gradients are applied on each patch. Next, the resulting histograms are concatenated, and their dimensionality is then reduced using Kernel Principle Component Analysis. Once completed, patches are randomly selected using the concept of random sampling to finally construct several sub-Support Vector Machine classifiers. The results obtained from these sub-classifiers are combined to generate the final recognition outcome. Experimental results based on the AR face database and the Extended Yale B database show the effectiveness of our proposed technique.

The image features of emotional faces that predict the initial eye movement to a face

Emotional facial expressions are important visual communication signals that indicate a sender’s intent and emotional state to an observer. As such, it is not surprising that reactions to different expressions are thought to be automatic and independent of awareness. What is surprising, is that studies show inconsistent results concerning such automatic reactions, particularly when using different face stimuli. We argue that automatic reactions to facial expressions can be better explained, and better understood, in terms of quantitative descriptions of their low-level image features rather than in terms of the emotional content (e.g. angry) of the expressions. Here, we focused on overall spatial frequency (SF) and localized Histograms of Oriented Gradients (HOG) features. We used machine learning classification to reveal the SF and HOG features that are sufficient for classification of the initial eye movement towards one out of two simultaneously presented faces. Interestingly, the identified features serve as better predictors than the emotional content of the expressions. We therefore propose that our modelling approach can further specify which visual features drive these and other behavioural effects related to emotional expressions, which can help solve the inconsistencies found in this line of research.

Multimodal biometrics of fingerprint and signature recognition using multi-level feature fusion and deep learning techniques

Providing security in biometrics is the major challenging task in the current situation. A lot of research work is going on in this area. Security can be more tightened by using complex security systems, like by using more than one biometric trait for recognition. In this paper multimodal biometric models are developed to improve the recognition rate of a person. The combination of physiological and behavioral biometrics characteristics is used in this work. Fingerprint and signature biometrics characteristics are used to develop a multimodal recognition system. Histograms of oriented gradients (HOG) features are extracted from biometric traits and for these feature fusions are applied at two levels. Features of fingerprint and signatures are fused using concatenation, sum, max, min, and product rule at multilevel stages, these features are used to train deep learning neural network model. In the proposed work, multi-level feature fusion for multimodal biometrics with a deep learning classifier is used and results are analyzed by a varying number of hidden neurons and hidden layers. Experiments are carried out on SDUMLA-HMT, machine learning and data mining lab, Shandong University fingerprint datasets, and MCYT signature biometric recognition group datasets, and encouraging results were obtained.

Computer vision model for detecting block falls at the martian north polar region.

<p>Dynamic changes of Martian north polar scarps present a valuable insight into the planet's natural climate cycles (Byrne, 2009; Head et al., 2003)<sup>1,2</sup>. Annual avalanches and block falls are amongst the most noticeable surface processes that can be directly linked with the extent of the latter dynamics (Fanara et al, 2020)<sup>3</sup>. New remote sensing approaches based on machine learning allow us to make precise records of the aforementioned mass wasting activity by automatically extracting and analyzing bulk information obtained from satellite imagery.  Previous studies have concluded that a Support Vector Machine (SVM) classifier trained using Histograms of Oriented Gradients (HOG) can be used to efficiently detect block falls, even against backgrounds with increased complexity (Fanara et al., 2020)<sup>4</sup>. We hypothesise that this pretrained model can now be utilized to generate an extended dataset of labelled image data, sufficient in size to opt for a deep learning approach. On top of improving the detection model we also attempt to address the image co-registration protocol. Prior research has suggested this to be a substantial bottleneck, which reduces the amounts of suitable images. We plan to overcome these limitations either by extending our model to include multi-sensor data, or by deploying improved methods designed for exclusively optical data (e.g.  COSI-CORR software (Ayoub, Leprince and Avouac, 2017)<sup>5</sup>).  The resulting algorithm should be a robust solution capable of improving on the already established baselines of 75.1% and 8.5% for TPR and FDR respectively (Fanara et al., 2020)4. The NPLD is our primary area of interest due to it’s high levels of activity and good satellite image density, yet we also plan to apply our pipeline to different surface changes and Martian regions as well as on other celestial objects.</p><p> </p><p>1. Head, J.W., Mustard, J.F., Kreslavsky, M.A., Milliken, R.E., Marchant, D.R., 2003. Recent ice ages on Mars. Nature 426, 797–802</p><p>2. Byrne, S., 2009. The polar deposits of Mars. Annu. Rev. Earth Planet. Sci. 37, 535–560.</p><p>3. Fanara, K. Gwinner, E. Hauber, J. Oberst, Present-day erosion rate of north polar scarps on Mars due to active mass wasting; Icarus,Volume 342, 2020; 113434, ISSN 0019-1035.</p><p>4. Fanara, K. Gwinner, E. Hauber, J. Oberst, Automated detection of block falls in the north polar region of Mars; Planetary and Space Science, Volume 180, 2020; 104733, ISSN 0032-0633.</p><p>5. Ayoub, F.; Leprince, S.; Avouac, J.-P. User’s Guide to COSI-CORR Co-registration of Optically Sensed Images and Correlation; California Institute of Technology: Pasadena, CA, USA, 2009; pp. 1–49.</p>