Phenetic Research Articles

Deep learning framework for geological symbol detection on geological maps

Dynamic legend generation for geological maps aims to detect and identify geological map symbols within the current viewshed and generate a corresponding real-time legend to help users quickly obtain the name and meaning of symbols. Detection and recognition entail high complexity and uncertainty because of the diversity of symbol types and the randomness of symbol distribution, and thus the generation of dynamic legends for geological maps is challenging. A new framework based on deep learning is proposed in this study, combining the deep convolutional neural network (CNN) and graph convolutional network (GCN) to realize the extraction and recognition of geological map symbols. Within the framework, a CNN-based model called single symbol detection network (SSDN) is developed to detect and identify single geological map symbols, and a novel GCN combined with L2 distance attention (DAGCN) is proposed to deal with the difficulty of extracting compound symbols caused by the randomness of symbol distribution. This work systematically solves the problem of geological symbol detection with the aid of object detection technology based on deep learning, providing foundation for the dynamic legend generation. Experiments show that the framework of the proposed method is effective, and a new benchmark is established for geological symbol detection on geological maps. All of our data and code are publicly available.

A comparison of two dissimilarity functions for mixed-type predictor variables in the delta -machine

The delta -machine is a statistical learning tool for classification based on dissimilarities or distances between profiles of the observations to profiles of a representation set, which was proposed by Yuan et al. (J Claasif 36(3): 442–470, 2019). So far, the delta -machine was restricted to continuous predictor variables only. In this article, we extend the delta -machine to handle continuous, ordinal, nominal, and binary predictor variables. We utilized a tailored dissimilarity function for mixed type variables which was defined by Gower. This measure has properties of a Manhattan distance. We develop, in a similar vein, a Euclidean dissimilarity function for mixed type variables. In simulation studies we compare the performance of the two dissimilarity functions and we compare the predictive performance of the delta -machine to logistic regression models. We generated data according to two population distributions where the type of predictor variables, the distribution of categorical variables, and the number of predictor variables was varied. The performance of the delta -machine using the two dissimilarity functions and different types of representation set was investigated. The simulation studies showed that the adjusted Euclidean dissimilarity function performed better than the adjusted Gower dissimilarity function; that the delta -machine outperformed logistic regression; and that for constructing the representation set, K-medoids clustering achieved fewer active exemplars than the one using K-means clustering while maintaining the accuracy. We also applied the delta -machine to an empirical example, discussed its interpretation in detail, and compared the classification performance with five other classification methods. The results showed that the delta -machine has a good balance between accuracy and interpretability.

An analytical formulation for multidimensional continuous opinion models

Usually, opinion formation models assume that individuals have an opinion about a given topic which can change due to interactions with others. However, individuals can have different opinions on different topics and therefore n-dimensional models are best suited to deal with these cases. While there have been many efforts to develop analytical models for one dimensional opinion models, less attention has been paid to multidimensional ones. In this work, we develop an analytical approach for multidimensional models of continuous opinions. We show that for any generic reciprocal interactions between agents, the mean value of initial opinion distribution is conserved. Moreover, for positive social influence interaction mechanisms, the variance of opinion distributions decreases with time and the system converges to a delta distributed function. In particular, we calculate the convergence time when agents get closer in a discrete quantity after interacting, showing a clear difference between cases where the approach is through Manhattan or Euclidean distance.

Frontal EEG Asymmetry in Major Depression, Dysthymia and Bipolar Disorder

Depression is a mental disorder that continues to make life difficult or impossible for a depressed person and, if left untreated, can lead to dangerous activities such as self-harm and suicide. Nowadays, Electroencephalogram (EEG) has become an important diagnostic tool for many brain disorders. In this article, a new method for the detection of depression based on the reconstructed phase space (RPS) of EEG signals and geometrical features has been proposed. The RPS of the EEG signals of 22 normal and 22 depressed subjects are plotted in two-dimensional space and, based on their shape, 34 geometrical features are extracted. The p-values for the proposed features were significantly lower (p-value ≈ 0) indicating the capacity of the proposed geometric features for the normal and depression EEG signals classification tasks. For the purpose of reducing feature vector arrays, the performance of four optimization algorithms is checked, namely: ant colony optimization (ACO), grey wolf optimization (GWO), genetic algorithm (GA) and particle swarm optimization (PSO), in which GA with the ability of 58.8% was better than the other optimization algorithms for decreasing the feature vector arrays. Selected features are fed to the support vector machine (SVM) classifier with radial basis function (RBF) kernel and K-nearest neighbors (KNN) classifier with Euclidean and city block distances in 10-fold cross-validation (CV) strategy. The proposed framework achieved a fairly good average classification accuracy (ACC) of 99.30% and a Matthews correlation coefficient (MCC) of 0.98 using the selected features of the PSO algorithm and the SVM classifier. We found that the RPS of normal EEG signals has a more irregular, complex and unpredictable shape than the RPS of depression EEG signals which has more regular (simple) with less variation and more predictable shape; therefore, we can say that RPS of EEG signals can be used as a biomarker for psychiatrists which are simpler than the EEG signals in visual depression diagnostics. We also found that EEG signals from the right hemisphere are significant for depression detection than the left hemisphere. The proposed framework may be used in clinics and hospitals to detect depression disorder quickly and precisely.

Image Mining Based on Deep Belief Neural Network and Feature Matching Approach Using Manhattan Distance

Over the past few decades multimedia content, particularly digital images, has increased at a rapid pace, with several complex images being uploaded to various social websites such as Instagram, Facebook and Twitter. Therefore, it is difficult to search and retrieve the relevant image in seconds. Search engines retrieve images based on traditional textbased methods that depend on metadata and captions. In the last few years, a wide range of research has focused on content-based image retrieval (CBIR) based on image mining approaches. This is a challenging research area due to the ever-increasing multimedia database and other image libraries. In order to offer an effective search and retrieval, a novel CBIR system is proposed using the image mining-based deep belief neural network (IMDBN) technique. The proposed method is designed to enhance retrieval accuracy while diminishing the semantic gap between human visual understanding and image feature representation. To achieve this objective, the proposed system carries several steps like preprocessing, feature extraction, classification, and feature matching. Initially, the input database images are fed into the proposed image mining-based CBIR system, whereas colour-shape-texture (CST) feature extraction technique is applied to extract relevant feature set. The extracted features are fused and stored in the feature vector and aresubjected to the proposed IMDBN classification step to retrieve similar images in one label. Whenever a new query content is created, the most relevant images are retrieved. This, in turn, achieves 94% accuracy, which is higher than in existing approaches.

Curve Registration of Functional Data for Approximate Bayesian Computation

Approximate Bayesian computation is a likelihood-free inference method which relies on comparing model realisations to observed data with informative distance measures. We obtain functional data that are not only subject to noise along their y axis but also to a random warping along their x axis, which we refer to as the time axis. Conventional distances on functions, such as the L2 distance, are not informative under these conditions. The Fisher–Rao metric, previously generalised from the space of probability distributions to the space of functions, is an ideal objective function for aligning one function to another by warping the time axis. We assess the usefulness of alignment with the Fisher–Rao metric for approximate Bayesian computation with four examples: two simulation examples, an example about passenger flow at an international airport, and an example of hydrological flow modelling. We find that the Fisher–Rao metric works well as the objective function to minimise for alignment; however, once the functions are aligned, it is not necessarily the most informative distance for inference. This means that likelihood-free inference may require two distances: one for alignment and one for parameter inference.

AN INTERPRETABLE MACHINE LEARNING MODEL FOR INDIVIDUALIZED PROTOCOL SELECTION AND GONADOTROPIN DOSE SELECTION DURING OVARIAN STIMULATION

To develop an interpretable machine learning model for individualized gonadotropin dose selection during controlled ovarian stimulation. Historical, de-identified electronic medical record (EMR) data was collected from 4 IVF clinics in the United States. Records were filtered for autologous, non-canceled IVF retrievals, resulting in 7,977 cycles started between 2014 and 2020. A multiple linear regression model was developed with cross validation and recursive feature elimination to predict the number of eggs and mature (MII) eggs retrieved using baseline parameters available prior to start of treatment. The predictor variables were then used to create a patient similarity model based on K nearest neighbors (KNN), an interpretable machine learning technique. After identifying the best performing distance metrics, neighbor weights, and number of neighbors, the model was used to predict the number of eggs and MII eggs retrieved by calculating the weighted average from the set of K neighbors most similar to the patient of interest. The performance of the KNN model was compared to linear regression in terms of R-squared (R2) and mean absolute error (MAE). The KNN model was then used to (a) query the K most similar patients, and (b) identify the optimal gonadotropin dose in terms of highest number of MII eggs retrieved. We developed linear regression and KNN models using patient age, BMI, diagnosis, AMH, AFC, number of previous IVF cycles, and parity. KNN achieved highest performance using the Manhattan distance, 50-80 similar patients, and distance-based neighbor weighting. The KNN model outperformed linear regression for eggs retrieved (R2: 0.43 vs. 0.39, MAE: 4.84 vs. 4.98) and for MII eggs retrieved (R2: 0.39 vs. 0.35, MAE: 4.01 vs. 4.11). We then investigated the application of these models for gonadotropin dose selection. Linear models indicated that gonadotropin dose is negatively correlated with MII eggs, which may in part reflect that poor-prognosis patients are prescribed higher doses. In contrast, the KNN model showed that 22% of patients had a concave dose response curve, in which there was an optimal dose that maximized the number of MII eggs. We developed a patient similarity model using K nearest neighbors. The model showed better accuracy than linear regression for predicting eggs and MII eggs retrieved, and allowed the evaluation of which starting dose maximized the number of MII eggs retrieved, which is not possible with a linear model. Future work will optimize techniques for matching similar patients and extend the modeling for protocol selection.

Training Affective Computer Vision Models by Crowdsourcing Soft-Target Labels.

Emotion detection classifiers traditionally predict discrete emotions. However, emotion expressions are often subjective, thus requiring a method to handle compound and ambiguous labels. We explore the feasibility of using crowdsourcing to acquire reliable soft-target labels and evaluate an emotion detection classifier trained with these labels. We hypothesize that training with labels that are representative of the diversity of human interpretation of an image will result in predictions that are similarly representative on a disjoint test set. We also hypothesize that crowdsourcing can generate distributions which mirror those generated in a lab setting. We center our study on the Child Affective Facial Expression (CAFE) dataset, a gold standard collection of images depicting pediatric facial expressions along with 100 human labels per image. To test the feasibility of crowdsourcing to generate these labels, we used Microworkers to acquire labels for 207 CAFE images. We evaluate both unfiltered workers as well as workers selected through a short crowd filtration process. We then train two versions of a ResNet-152 neural network on soft-target CAFE labels using the original 100 annotations provided with the dataset: (1) a classifier trained with traditional one-hot encoded labels, and (2) a classifier trained with vector labels representing the distribution of CAFE annotator responses. We compare the resulting softmax output distributions of the two classifiers with a 2-sample independent t-test of L1 distances between the classifier's output probability distribution and the distribution of human labels. While agreement with CAFE is weak for unfiltered crowd workers, the filtered crowd agree with the CAFE labels 100% of the time for happy, neutral, sad and "fear + surprise", and 88.8% for "anger + disgust". While the F1-score for a one-hot encoded classifier is much higher (94.33% vs. 78.68%) with respect to the ground truth CAFE labels, the output probability vector of the crowd-trained classifier more closely resembles the distribution of human labels (t=3.2827, p=0.0014). For many applications of affective computing, reporting an emotion probability distribution that accounts for the subjectivity of human interpretation can be more useful than an absolute label. Crowdsourcing, including a sufficient filtering mechanism for selecting reliable crowd workers, is a feasible solution for acquiring soft-target labels.

Score-based likelihood ratios for linguistic text evidence with a bag-of-words model

The likelihood ratio paradigm for quantifying the strength of evidence has been researched in many fields of forensic science. Within this paradigm, score-based approaches for estimating likelihood ratios are becoming more prevalent in the forensic science literature. In this study, a score-based approach for estimating likelihood ratios is implemented for linguistic text evidence. Text data are represented via a bag-of-words model with the Z-score normalised relative frequencies of selected most-frequent words (the number of the most-frequent words = N), and the Euclidean, Manhattan and Cosine distance measures are trialled as the score-generating functions for comparing paired text samples. The score-to-likelihood-ratio conversion model was built using a common source method, and the best fitting model was selected from the parametric models of the Normal, Log-normal, Gamma and Weibull distributions. With the Amazon Product Data Authorship Verification Corpus, two groups of documents (each group including documents of approximately 700, 1400 and 2100 words) were synthesised for each author, allowing 720 same-author comparisons and 517,680 different-author comparisons to test the validity of the system. A series of experiments was conducted using combinations of the following conditions: the three score functions, the different values of N for the feature vector and the different document lengths. The validity of the system was assessed using the log-likelihood-ratio cost (Cllr), and the strength of the derived likelihood ratios was charted in the form of Tippett plots. It was demonstrated that 1) the Cosine measure consistently outperforms the other measures—the best performance is achieved with N = 260, regardless of the document length (e.g., Cllr values of 0.70640, 0.45314 and 0.30692, respectively, for 700, 1400 and 2100 words)—and 2) the derived likelihood ratios are very well calibrated irrespective of the distance measures and document lengths. A follow-up experiment showed that the described score-based approach is relatively robust and stable for a limited quantity of background data. The derived likelihood ratios that were estimated separately to the three distance measures were logistic regression fused; and the fusion achieved a further improvement in performance—for example, a Cllr of 0.23494 for 2100 words. This study demonstrates the possibility of designing likelihood ratio–based systems that discriminate between same-author and different-author documents.

Fast Path Planning Through Segmentation of the Map into Manhattan-Cohesive Areas

This paper presents a method for fast planning within arbitrary maps, through segmentation of the map into Manhattan-cohesive areas. A Manhattan-cohesive area is a connected part of the map where the optimal distance between any two points in the area is equal to their Manhattan distance. We adopt a four directions Manhattan distance, where diagonal moves are allowed. In the paper we present the method we adopted to fragmentize the map, as well the method to extract the paths. The proposed method produces nearly optimal plans quite efficiently.

Implementing Fast Heuristic Search Code

Published papers rarely disclose implementation details. In this paper we show how such details can account for speedups of up to a factor of 28 for different implementations of the same algorithm. We perform an in-depth analysis of the most popular benchmark in heuristic search: the 15-puzzle. We study implementation choices in C++ for both IDA* and A* using the Manhattan distance heuristic. Results suggest that several optimizations deemed critical in folklore provide only small improvements while seemingly innocuous choices can play a large role. These results are important for ensuring that the correct conclusions are drawn from empirical comparisons

Deep Contrast Learning Approach for Address Semantic Matching

Address is a structured description used to identify a specific place or point of interest, and it provides an effective way to locate people or objects. The standardization of Chinese place name and address occupies an important position in the construction of a smart city. Traditional address specification technology often adopts methods based on text similarity or rule bases, which cannot handle complex, missing, and redundant address information well. This paper transforms the task of address standardization into calculating the similarity of address pairs, and proposes a contrast learning address matching model based on the attention-Bi-LSTM-CNN network (ABLC). First of all, ABLC use the Trie syntax tree algorithm to extract Chinese address elements. Next, based on the basic idea of contrast learning, a hybrid neural network is applied to learn the semantic information in the address. Finally, Manhattan distance is calculated as the similarity of the two addresses. Experiments on the self-constructed dataset with data augmentation demonstrate that the proposed model has better stability and performance compared with other baselines.

An assessment of the effect of using different mappings and Minkowski distances in joint monitoring of the time-between-event processes

Monitoring multiple parameters of a process using a single integrated charting scheme is an attractive research area in statistical process monitoring and control. The Max-type combining function based on Chebyshev’s distance and the Distance-type combining function, essentially based on Euclidean distance, are widely used to construct various schemes for simultaneously monitoring the location and scale parameters. In most of these schemes, normalising the suitable function of maximum likelihood estimators (MLE) of individual parameters is commonly used. While monitoring two-parameter exponential distributions, we show that mapping the pivots based on the maximum likelihood estimators to standard normal variables is not optimal. This paper investigates four different mappings to analyse the transformation effect on the joint monitoring schemes for a two-parameter exponentially distributed process. The Chebyshev’s and Euclidean distances are particular cases of Minkowski distance. We additionally investigate the effect of Manhattan and minimum-type Minkowski distances via Monte Carlo in terms of the run-length properties. The overall chart performance is assessed using the expected weighted run length EWRL. It is observed that the use of Manhattan distance and mapping into the standard Laplace model is more appropriate. A real example of monitoring a high-voltage current in a P-type high-voltage metal oxide semiconductor transistor (HPM) data is given to show the excellent performance of the suggested control chart.

Derived vectorcardiogram based automated detection of posterior myocardial infarction using FBSE-EWT technique

The early detection of posterior myocardial infarction (PMI) is an important task as it can cause cardiac failure. Due to the absence of extra posterior leads in the standard 12-lead electrocardiogram (ECG), the PMI detection sensitivity degrades. To improve it, additional posterior leads can be included in standard 12-lead ECG system. Another approach utilizing vectorcardiogram (VCG) has been in practice with specific 7 electrodes with one posterior lead. Although the VCG approach is promising, the arrangement of posterior lead causes patient discomfort. To overcome aforesaid issue, derived VCG (dVCG) obtained through the Dowers inverse transformation has been proposed, wherein, from the standard 12-lead ECG, a three lead VCG is derived, which then can be utilized for PMI detection. Relying on the dVCG approach, in the present paper, we have proposed a novel methodology for PMI detection using Fourier-Bessel series expansion based empirical wavelet transform (FBSE-EWT). The non-stationary behavior of dVCG has been captured using FBSE-EWT, followed by the principal component analysis employing eigenvalues of covariance matrix. The constructed feature space is then fed to decision tree (DT), support vector machine (SVM), and K-nearest neighbor (KNN) classifiers. The KNN classifier with inverse squared city block distance resulted in the overall classification accuracy of 97.92% with sensitivity and specificity of 96.63% and 98.50%, respectively. The experiments were performed over Physikalisch-Technische Bundesanstalt Diagnostic (PTBD) dataset. The proposed method has the potential to be utilized in accurate and robust PMI detection from dVCG signals in the clinical settings.

Dispersing and grouping points on planar segments

Motivated by (continuous) facility location, we study the problem of dispersing and grouping points on a set of segments (of streets) in the plane. In the former problem, given a set of n disjoint line segments in the plane, we investigate the problem of computing a point on each of the n segments such that the minimum Euclidean distance between any two of these points is maximized. We prove that this 2D dispersion problem is NP-hard, in fact, it is NP-hard even if all the segments are parallel and are of unit length. This is in contrast to the polynomial solvability of the corresponding 1D problem by Li and Wang (2016), where the intervals are in 1D and are all disjoint. With this result, we also show that the Independent Set problem on Colored Linear Unit Disk Graph (meaning the convex hulls of points with the same color form disjoint line segments) remains NP-hard, and the parameterized version of it is in W[2]. In the latter problem, given a set of n disjoint line segments in the plane we study the problem of computing a point on each of the n segments such that the maximum Euclidean distance between any two of these points is minimized. We present a factor-1.1547 approximation algorithm which runs in O(nlog⁡n) time. Our results can be generalized to the Manhattan distance.

Globally Optimal Facility Locations for Continuous-Space Facility Location Problems

The continuous-space single- and multi-facility location problem has attracted much attention in previous studies. This study focuses on determining the globally optimal facility locations for two- and higher-dimensional continuous-space facility location problems when the Manhattan distance is considered. Before we propose the exact method, we start with the continuous-space single-facility location problem and obtain the global minimizer for the problem using a statistical approach. Then, an exact method is developed to determine the globally optimal solution for the two- and higher-dimensional continuous-space facility location problem, which is different from the previous clustering algorithms. Based on the newly investigated properties of the minimizer, we extend it to multi-facility problems and transfer the continuous-space facility location problem to the discrete-space location problem. To illustrate the effectiveness and efficiency of the proposed method, several instances from a benchmark are provided to compare the performances of different methods, which illustrates the superiority of the proposed exact method in the decision-making of the continuous-space facility location problems.

New Approaches in Characterization of Lesions Dissemination in DLBCL Patients on Baseline PET/CT.

Simple SummaryRecently, a new PET parameter expressing lymphoma dissemination has been proposed to identify high-risk DLBCL patients: the distance between the two furthest lesions, standardized by body surface area (SDmax). This study aimed to determine the best way to measure the distance between lesions, by comparing different methods of distance measurements. We obtained similar results in terms of prediction of outcome between the different methods further validating the relevance of the dissemination features. We highlighted the possibility to calculate it directly from lymphoma voxels instead of lesion centroids, and thus applied it to a metabolic tumor volume (MTV) determined by deep learning algorithms. This could allow the use in clinical practice of this parameter, characterizing tumor spread, in combination with the tumor burden, for patient risk stratification. Dissemination, expressed recently by the largest Euclidian distance between lymphoma sites (SDmax), appeared a promising risk factor in DLBCL patients. We investigated alternative distance metrics to characterize the robustness of the dissemination information. In 290 patients from the REMARC trial (NCT01122472), the Euclidean (Euc), Manhattan (Man), and Tchebychev (Tch) distances between the furthest lesions, firstly based on the centroid of each lesion and then directly from the two most distant tumor voxels and the Travelling Salesman Problem distance (TSP) were calculated. For PFS, the areas under the ROC curves were between 0.63 and 0.64, and between 0.62 and 0.65 for OS. Patients with high SDmax whatever the method of calculation or high SD_TSP had a significantly poorer outcome than patients with low SDmax or SD_TSP (p < 0.001 for both PFS and OS), with significance maintained in Ann Arbor advanced-stage patients. In multivariate analysis with total metabolic tumor volume and ECOG, each distance feature had an independent prognostic value for PFS. For OS, only SDmax_Tch, SDmax_Euc _Vox, and SDmax_Man _Vox reached significance. The spread of DLBCL lesions measured by the largest distance between lymphoma sites is a strong independent prognostic factor and could be measured directly from tumor voxels, allowing its development in the area of the deep learning segmentation methods.

Sensitivity of shortest distance search in the ant colony algorithm with varying normalized distance formulas

The ant colony algorithm is an algorithm adopted from the behavior of ants which naturally ants are able to find the shortest route on the way from the nest to places of food sources based on footprints on the track that has been passed. The ant colony algorithm helps a lot in solving several problems such as scheduling, traveling salesman problems (TSP) and vehicle routing problems (VRP). In addition, ant colony has been developed and has several variants. However, in its function to find the shortest distance is optimized by utilizing several normalized distance formulas with the data used in finding distances between merchants in the mercant ecosystem. Where in the test normalized distance is superior Hamming distance in finding the shortest distance of 0.2875, then followed by the same value, namely the normalized formula Manhattan distance and normalized Euclidean distance with a value of 0.4675 and without using the normalized distance formula or the original ant colony algorithm gets a value 0.6635. Given the sensitivity in distance search using merchant ecosystem data, the method works well on the ant colony Algorithm using normalized Hamming distance.

Calculation of Signaling Block Distances in High Speed ​​Train Lines

Raylı ulaşım sistemlerinin hızlı, güvenilir ve emniyetli olabilmesi için sinyalizasyon sistemlerine ihtiyaç vardır. Sinyalizasyon sistemleri geçmişten günümüze birçok değişikliğe uğramış olmasına rağmen temel prensipler her zaman aynı kalmıştır. Kazaların, can ve mal kayıplarının önüne geçebilmek, güvenli işletme sağlayabilmek için trenler arasında belirli bir zaman ve mesafenin olması gerekmektedir. Bu çalışmada yüksek hızlı tren hatlarında sinyal blok mesafeleri hesaplanarak, Ankara-Eskişehir ile Ankara-Konya YHT (Yüksek Hızlı Tren) hatlarındaki blok mesafeleri ile karşılaştırmaları yapılmıştır. Sonuç bölümünde blok mesafelerinin hat kapasitesine etkisi, kurulum ve bakım maliyetleri değerlendirilerek tasarım ve uygulama aşamalarına dair önerilerde bulunulmuştur.

A Study of Deep Learning-Based Face Recognition Models for Sibling Identification.

Accurate identification of siblings through face recognition is a challenging task. This is predominantly because of the high degree of similarities among the faces of siblings. In this study, we investigate the use of state-of-the-art deep learning face recognition models to evaluate their capacity for discrimination between sibling faces using various similarity indices. The specific models examined for this purpose are FaceNet, VGGFace, VGG16, and VGG19. For each pair of images provided, the embeddings have been calculated using the chosen deep learning model. Five standard similarity measures, namely, cosine similarity, Euclidean distance, structured similarity, Manhattan distance, and Minkowski distance, are used to classify images looking for their identity on the threshold defined for each of the similarity measures. The accuracy, precision, and misclassification rate of each model are calculated using standard confusion matrices. Four different experimental datasets for full-frontal-face, eyes, nose, and forehead of sibling pairs are constructed using publicly available HQf subset of the SiblingDB database. The experimental results show that the accuracy of the chosen deep learning models to distinguish siblings based on the full-frontal-face and cropped face areas vary based on the face area compared. It is observed that VGGFace is best while comparing the full-frontal-face and eyes—the accuracy of classification being with more than 95% in this case. However, its accuracy degrades significantly when the noses are compared, while FaceNet provides the best result for classification based on the nose. Similarly, VGG16 and VGG19 are not the best models for classification using the eyes, but these models provide favorable results when foreheads are compared.