Frontal Face Research Articles

Upstream velocity fields induced by frontal jet injection in a square cylinder

The flow characteristics around the frontal face of a square cylinder with a frontal jet injection were experimentally studied in a closed-loop wind tunnel using the PIV technique. Four characteristic flow modes were identified based on the injection ratio: swinging jet, deflected oscillating jet, deflection jet, and penetrating jet. At low injection ratio, the flow mode denoted as the swinging jet is characterised by two recirculation regions and a four-way saddle point. At moderate injection ratio, the flow mode termed deflected oscillating jet is characterised by a recirculation region and a counterclockwise rotating vortex within the recirculation region. At moderately high injection ratio, the flow mode is denoted as deflection jet, where a clockwise rotating vortex appears in the flow field. At high injection ratio, the flow mode is termed as penetrating jet, with no vortex formation in the flow field. The strength of the turbulence intensities increases considerably with high injection ratios. The time histories of instantaneous velocity for non-injection case, swinging jet, and deflected oscillating jet modes show periodic oscillations, whereas those of deflection jet and penetrating jet modes do not show periodic oscillations. The study presents and discusses the time-averaged velocity vectors, and streamlines, vorticity contours, velocity properties, velocity time histories, and power spectral density function under different flow modes.

Deep learning predicted perceived age is a reliable approach for analysis of facial ageing: A proof of principle study.

Perceived age (PA) has been associated with mortality, genetic variants linked to ageing and several age-related morbidities. However, estimating PA in large datasets is laborious and costly to generate, limiting its practical applicability. To determine if estimating PA using deep learning-based algorithms results in the same associations with morbidities and genetic variants as human-estimated perceived age. Self-supervised learning (SSL) and deep feature transfer (DFT) deep learning (DL) approaches were trained and tested on human-estimated PAs and their corresponding frontal face images of middle-aged to elderly Dutch participants (n = 2679) from a population-based study in the Netherlands. We compared the DL-estimated PAs with morbidities previously associated with human-estimated PA as well as genetic variants in the gene MC1R; we additionally tested the PA associations with MC1R in a new validation cohort (n = 1158). The DL approaches predicted PA in this population with a mean absolute error of 2.84 years (DFT) and 2.39 years (SSL). In the training-test dataset, we found the same significant (p < 0.05) associations for DL PA with osteoporosis, ARHL, cognition, COPD and cataracts and MC1R, as with human PA. We also found a similar but less significant association for SSL and DFT PAs (0.69 and 0.71 years per allele, p = 0.008 and 0.011, respectively) with MC1R variants in the validation dataset as that found with human, SSL and DFT PAs in the training-test dataset (0.79, 0.78 and 0.71 years per allele respectively; all p < 0.0001). Deep learning methods can automatically estimate PA from facial images with enough accuracy to replicate known links between human-estimated perceived age and several age-related morbidities. Furthermore, DL predicted perceived age associated with MC1R gene variants in a validation cohort. Hence, such DL PA techniques may be used instead of human estimations in perceived age studies thereby reducing time and costs.

Validity of facial skin analysis pore detection: A comparative analysis.

Reliable, objective measures to assess facial characteristics would aid in the assessment of many dermatological treatments. Previous work utilized an iOS application-based artificial intelligence (AI) tool compared to the "gold standard" computer-based and a physician assessment on five skin metrics (British Journal of Dermatology, 2013, 169, 474). The AI tool had superior agreement for all skin metrics except pores and subsequently underwent an algorithm update for its pore detection system. This comparative analysis assessed the performance of the updated AI tool's pore scores across all Fitzpatrick skin phototypes to determine whether the AI tool more accurately represents a dermatologist's assessment of pores. Frontal facing photographs in uniform lighting conditions were taken of each participant. Percentile scores were generated by each of the four self-learning models of the updated AI tool. The pore percentile scores generated by the original and updated AI tool were used to rate "worse" pores among participant pairs. These ratings were compared to pore assessments performed by a "gold-standard" device and a board-certified dermatologist. Compared to the original pore detection tool and the computer-based program, models A and D had the highest concordance with the physician's pore assessments for Fitzpatrick skin phototypes III-IV and V-VI, respectively. The AI tool's pores detection update was successful in its ability to accurately detect pores on all Fitzpatrick skin types, improving on the performance of the AI prior to the update. Responsibly developed AI tools that can accurately and reliably detect skin metrics across diverse Fitzpatrick skin types can facilitate dermatologic evaluation, individualize treatment, and determine treatment response.

DSU-GAN: A robust frontal face recognition approach based on generative adversarial network

Face recognition technology is widely used in different areas, such as entrance guard, payment etc. However, little attention has been given to non-positive faces recognition, especially model training and the quality of the generated images. To this end, a novel robust frontal face recognition approach based on generative adversarial network (DSU-GAN) is proposed in this paper. A mechanism of consistency loss is presented in deformable convolution proposed in the generator-encoder to avoid additional computational overhead and the problem of overfitting. In addition, a self-attention mechanism is presented in generator–encoder to avoid information overloading and construct the long-term dependencies at the pixel level. To balance the capability between the generator and discriminator, a novelf discriminator architecture based U-Net is proposed. Finally, the single-way discriminator is improved through a new up-sampling module. Experiment results demonstrate that our proposal achieves an average Rank-1 recognition rate of 95.14% on the Multi-PIE face dataset in dealing with the multi-pose. In addition, it is proven that our proposal has achieved outstanding performance in recent benchmarks conducted on both IJB-A and IJB-C.

Stress Detection through Compound Facial Expressions Using Neural Networks

Human face is the most dynamic part of the body that conveys information about the instant emotions. Facial expression analysis starts from early 1900s where later on scientists identify the six basic facial expressions as Anger, Disgust, Fear, Happiness, Sadness, Surprise and Neutral with the pioneering studies of psychologists. In the last decades, the acceleration in artificial intelligence and computer vision research makes it possible to automatically detect facial expressions through images. Furthermore, micro expressions, muscle movements and compound facial expressions; that are the combinations of the basic expressions can be also analyzed with computer vision algorithms. The main motivation in automatic facial expression analysis is to support human-computer. Furthermore, facial expression analysis can be a driver for automatic emotion analysis. In this study, we propose a novel method to detect stress indicators on the frontal face images. The detection procedure is based on compound facial expression analysis. 49 couples of 6 basic facial expressions where one is dominating, and the other is the complementary expression are employed. iCV-MEFED facial expression dataset is used in the experiments where video and image samples are provided for every compound facial expression class. The training and testing of compound facial expressions are done using a deep neural network. The robust representations of faces are achieved using a fusion method that combines deep texture features and the action units on the face. Then, through the appropriate grouping of the compound expressions, the system can detect the signs of stress. The proposed approach obtains encouraging results, and it is open to further improvements.

A Multimodal Dataset for Mixed Emotion Recognition

Mixed emotions have attracted increasing interest recently, but existing datasets rarely focus on mixed emotion recognition from multimodal signals, hindering the affective computing of mixed emotions. On this basis, we present a multimodal dataset with four kinds of signals recorded while watching mixed and non-mixed emotion videos. To ensure effective emotion induction, we first implemented a rule-based video filtering step to select the videos that could elicit stronger positive, negative, and mixed emotions. Then, an experiment with 80 participants was conducted, in which the data of EEG, GSR, PPG, and frontal face videos were recorded while they watched the selected video clips. We also recorded the subjective emotional rating on PANAS, VAD, and amusement-disgust dimensions. In total, the dataset consists of multimodal signal data and self-assessment data from 73 participants. We also present technical validations for emotion induction and mixed emotion classification from physiological signals and face videos. The average accuracy of the 3-class classification (i.e., positive, negative, and mixed) can reach 80.96% when using SVM and features from all modalities, which indicates the possibility of identifying mixed emotional states.

Accuracy of geometric morphometrics for age estimation using frontal face photographs of children and adolescents: A promising method for forensic practice

Age estimation is crucial in legal and humanitarian contexts. Forensic professionals may use various procedures to estimate age, including dental analysis, bone density tests, evaluation of physical characteristics including facial bone structure and development, and image-based methods. Although images are often the only material available, visual observation of photographic material is an imprecise method in age estimation, which can compromise judicial decision-making. Analyzing 4000 photographs from the Brazilian Federal Police database, representing four age groups (6, 10, 14, and 18 years), the study employed automated analysis by marking 28 photogrammetric points. Data were used to establish facial patterns by age and sex using the facial geometric morphometrics method. Performance was assessed through a Multinomial Logistic Regression model, evaluating accuracy, sensitivity, and specificity across the categorical age groups. Analyses were conducted using R software, with a 5 % significance level. The study found that facial geometric morphometrics achieved an overall accuracy of 69.3 % in age discrimination, with higher accuracy in males (74.7 %) compared to females (65.8 %) (p < 0.001). The method excelled at predicting the age of 6-year-olds with 87.3 % sensitivity and 95.6 % specificity but had lower performance at 14 years. It showed greater accuracy in distinguishing age groups with larger age gaps, achieving up to 99.5 % accuracy between certain groups, and was particularly effective in differentiating ages of 6 and 10 years in females and 10, 14, and 18 years in males. The facial geometric morphometrics emerges as a promising approach for age estimation among children and adolescents in forensic settings.

An efficient digital image forgery detection using Pelican search optimisation-based DCNN

ABSTRACT In the digital age, images are widely used to document events, provide evidence, and communicate information. Ensuring the authenticity of digital images is crucial to maintaining trust and integrity in various domains, including journalism, forensics, legal proceedings, and historical documentation. In this research, a PeCA-DCNN model for digital image detection is proposed, employing a hybrid approach. Initially, the data from image forgery databases is collected and preprocessed to eliminate noise and artefacts. Subsequently, the Viola–Jones algorithm detects frontal faces, and feature extraction is performed using pre-trained models VGG-16 and Resnet-101. To reduce the computational overhead, feature extraction is performed and generates a feature vector. The PeCA algorithm, combined with an adaptive self-boosted DCNN, is used to classify fake and genuine images. The PeCA algorithm enhances model performance by adjusting classifier parameters’ weights and biases. When evaluating the PeCA-DCNN, significant improvements in accuracy, sensitivity, and specificity are obtained with enhancement rates of 1.48, 3.06, and 0.05 in an 80% training scenario, and 3.92, 3.24, and 2.22 in k-fold cross-validation. These results demonstrate the effectiveness of the proposed approach compared to existing techniques.

An algorithm for two-dimensional pattern detection by combining Echo State Network-based weak classifiers

Pattern detection is one of the essential technologies in computer vision. To solve pattern detection problems, the system needs a vast amount of computational resources. To train a multilayer perceptron or convolutional neural network, the gradient descent method is commonly used. The method consumes computational resources. To reduce the amount of computation, we propose a two-dimensional pattern detection algorithm based on Echo State Network (ESN). The training rule of ESN is based on one-shot ridge regression, which enables us to avoid the gradient descent. ESN is a kind of recurrent neural network (RNN), which is often used to embed temporal signals inside the network, rarely used for the embedding of static patterns. In our prior work (Kage, 2023), we found that static patterns can be embedded in an ESN network by associating the training patterns with its stable states, or attractors. By using the same training procedure as our prior work, we made sure that we can associate each training patch image with the desired output vector. The resulting performance of a single ESN classifier is, however, relatively poor. To overcome this poor performance, we introduced an ensemble learning framework by combining multiple ESN weak classifiers. To evaluate the performance, we used CMU-MIT frontal face images (CMU DB). We trained eleven ESN-based classifiers by using six CMU DB training images and evaluated the performance by using a CMU DB test image. We succeeded in reducing false positives in the CMU DB test image down to 0.0515 %.

Depression recognition over fusion of visual and vocal expression using artificial intelligence

&lt;div class="WordSection1"&gt;&lt;p&gt;Depression is a mental illness that usually goes untreated in people and can have catastrophic consequences, including suicidal thoughts. Counselling services are widely available, but because depression is a stigmatized illness, many people who are depressed decide not to seek help. Therefore, it is essential to develop an automated system that can recognize depression in individuals before it worsens. In this study, a novel approach is proposed for identifying depression using a combination of visual and vocal emotions. Long short-term memory (LSTM) is used to assess verbal expressions and convolutional neural networks (CNN) to analyze facial expressions. The proposed system is trained using features of depression from the distress analysis interview corpus (DAIC) dataset and tested on videos of college students with frontal faces. The proposed approach is effective in detecting depression in individuals, with high accuracy and reliability.&lt;/p&gt;&lt;/div&gt;

Enhancing Driver Safety: Deep Learning Approach for Drowsiness Detection and Accident Prevention

Road safety concerns have spurred the development of innovative technologies aimed at reducing accidents, particularly those caused by driver fatigue. The main scope of the driver drowsiness detection system is to minimize road accidents caused by fatigue or sleepiness of drivers. This system leverages deep learning and computer vision, employing a Raspberry Pi camera to monitor facial expressions, including yawning, to assess the driver's alertness. Upon detecting signs of drowsiness, such as prolonged eye closure or altered facial expressions, the system triggers a buzzer alert by analyzing the Eye Aspect Ratio (EAR) and Mouth Aspect Ratio (MAR) ratios. Integrated within an embedded system, it utilizes frontal face detection algorithms and Haar cascade classifiers to localize key facial features in real-time, facilitating efficient monitoring for signs of fatigue or distraction. Additionally, it issues vibration alerts if the seat belt is not fastened. The seat belt remainder will be helpful to ensure the safety of the driver, reduces accidents and monitors the driver's heart rate, triggering a PANIC message if irregularities are detected. At traffic signals, the system automatically reduces vehicle speed using a vibration motor. In the unfortunate event of an accident, the system initiates speed reduction and promptly notifies registered contacts through the Blynk application, thereby significantly reducing accidents caused by drowsy driving.

De-occlusion and recognition of frontal face images: a comparative study of multiple imputation methods

AbstractIncreasingly, automatic face recognition algorithms have become necessary with the development and extensive use of face recognition technology, particularly in the era of machine learning and artificial intelligence. However, the presence of unconstrained environmental conditions degrades the quality of acquired face images and may deteriorate the performance of many classical face recognition algorithms. Due to this backdrop, many researchers have given considerable attention to image restoration and enhancement mechanisms, but with minimal focus on occlusion-related and multiple-constrained problems. Although occlusion robust face recognition modules, via sparse representation have been explored, they require a large number of features to achieve correct computations and to maximize robustness to occlusions. Therefore, such an approach may become deficient in the presence of random occlusions of relatively moderate magnitude. This study assesses the robustness of Principal Component Analysis and Singular Value Decomposition using Discrete Wavelet Transformation for preprocessing and city block distance for classification (DWT-PCA/SVD-L1) face recognition module to image degradations due to random occlusions of varying magnitudes (10% and 20%) in test images acquired with varying expressions. Numerical evaluation of the performance of the DWT-PCA/SVD-L1 face recognition module showed that the use of the de-occluded faces for recognition enhanced significantly the performance of the study recognition module at each level (10% and 20%) of occlusion. The algorithm attained the highest recognition rate of 85.94% and 78.65% at 10% and 20% occlusions respectively, when the MICE de-occluded face images were used for recognition. With the exception of Entropy where MICE de-occluded face images attained the highest average value, the MICE and RegEM result in images of similar quality as measured by their Absolute mean brightness error (AMBE) and peak signal to noise ratio (PSNR). The study therefore recommends MICE as a suitable imputation mechanism for de-occlusion of face images acquired under varying expressions.

Visual Field Restriction in the Recognition of Basic Facial Expressions: A Combined Eye Tracking and Gaze Contingency Study.

Uncertainties and discrepant results in identifying crucial areas for emotional facial expression recognition may stem from the eye tracking data analysis methods used. Many studies employ parameters of analysis that predominantly prioritize the examination of the foveal vision angle, ignoring the potential influences of simultaneous parafoveal and peripheral information. To explore the possible underlying causes of these discrepancies, we investigated the role of the visual field aperture in emotional facial expression recognition with 163 volunteers randomly assigned to three groups: no visual restriction (NVR), parafoveal and foveal vision (PFFV), and foveal vision (FV). Employing eye tracking and gaze contingency, we collected visual inspection and judgment data over 30 frontal face images, equally distributed among five emotions. Raw eye tracking data underwent Eye Movements Metrics and Visualizations (EyeMMV) processing. Accordingly, the visual inspection time, number of fixations, and fixation duration increased with the visual field restriction. Nevertheless, the accuracy showed significant differences among the NVR/FV and PFFV/FV groups, despite there being no difference in NVR/PFFV. The findings underscore the impact of specific visual field areas on facial expression recognition, highlighting the importance of parafoveal vision. The results suggest that eye tracking data analysis methods should incorporate projection angles extending to at least the parafoveal level.

Anthropometric Analysis of the Faces of Chinese Women in Leadership Roles.

The aim of this study was to investigate the appearance of the faces of Chinese women in leadership positions through an anthropometric analysis of Women of China from 1949 to 2022. Photographs of cover women of Women of China were gathered and a total of 380 frontal and 101 lateral photographs were collected. The photographs were divided into 3 time phases: phase I (1949-1978): from the launch of Women of China magazine, up until the time of reform and opening; phase II (1978-2010): from the time of reform and opening until when Google China was blocked; and phase III (2010-2022): to the present. The horizontal corneal diameter (11.52mm) was set as the reference value for the frontal face. The anthropometric results showed the width of the face, midface, and lower face decreased over time. The women at phase I had larger and round faces, smaller and narrower eyes, wider noses, and thicker lips; relatively masculine and hard-working woman image, as the occupations of women were mainly described as workers, farmers, and civil servants. Phase II showed smaller but still round faces, relatively wide-set round eyes, narrower noses, and smaller lips than in phase I; neutral features as skilled workers, engineers, entrepreneurs, civil servants, and researchers. Phase III showed oval faces, almond eyes, narrower noses, smaller full lips, and wider chins, reflecting a process of feminization; working in entertainment, sports, and science. The data from this study might serve as the recommended aesthetic proportions when performing plastic surgical procedures for talented woman.

Experimental Study of Wind Pressures on Low-Rise H-Shaped Buildings

Recognizing the urgent need for mitigating global warming, natural ventilation presents a potential strategy to reduce cooling energy demands, enhance thermal comfort, and contribute to indoor air quality. H-shaped buildings are prevalent worldwide, and they constitute the majority of the social housing construction in Brazil. Research suggests that the inadequate design of these buildings can result in poor ventilation; however, investigations about their natural ventilation performance are limited. Thus, the present contribution aims to determine the impact of the geometric characteristics of H-shaped buildings on the pressure distribution through wind tunnel experiments. Three models were tested in the wind tunnel experiments, representing different proportions. Their scales were configured to comply with the 5% obstruction limit allowed for wind tunnel testing, which was performed for 20 wind attack angles. Moreover, a scour test was carried out to allow a better understanding of the wind flow. Python scripting was developed to automate data processing, which is openly available in this paper. The results indicate that the proportion of the model influences the pressure distribution on roofs and leeward walls. Additionally, the depth of the recessed cavity affects its side surfaces and can result in a mirrored behavior on the frontal face of deep cavities (i.e., the wind direction is 45°). The model height influences the windward surfaces in its lower portion, since taller models present a recirculation vortex that modifies the pressure near the ground.

Head3D: Complete 3D Head Generation via Tri-plane Feature Distillation

Head generation with diverse identities is an important task in computer vision and computer graphics, widely used in multimedia applications. However, current full-head generation methods require a large number of three-dimensional (3D) scans or multi-view images to train the model, resulting in expensive data acquisition costs. To address this issue, we propose Head3D, a method to generate full 3D heads with limited multi-view images. Specifically, our approach first extracts facial priors represented by tri-planes learned in EG3D, a 3D-aware generative model, and then proposes feature distillation to deliver the 3D frontal faces within complete heads without compromising head integrity. To mitigate the domain gap between the face and head models, we present a dual-discriminator to guide the frontal and back head generation. Our model achieves cost-efficient and diverse complete head generation with photo-realistic renderings and high-quality geometry representations. Extensive experiments demonstrate the effectiveness of our proposed Head3D, both qualitatively and quantitatively.

Protecting occupant in autonomous vehicle accident with rotating seat and belt airbag

In autonomous vehicle, since that occupants no long need to operate the vehicle, they will have time to relax and entertain or communicate with other people. In order to improve the collision safety of the occupants who move freely and have different seat orientation angles in autonomous vehicle, a protection strategy based on the cooperative control of the rotating seat and belt airbag to avoid occupant injuries is proposed: in the pre-crash phase, the occupant on a rotated seat is repositioned into a close-to-standard frontal facing position, while the belt airbag deploys; in the crash phase, the fully deployed belt airbag cushions and absorbs occupant impact with the support of the occupant’s legs to reduce injury. Firstly, the current prediction time for unavoidable collisions in self-driving cars is about 350 ms, therefore, the airbag can be inflated by enlarging the inflation duration to reduce its impact during deployment. Due to the fixed installation position of the dashboard airbag, which cannot accompany the deficiency of seat movement, a new type of belt airbag is designed; Then, the simulation models of ±45° and ±90° seat orientation were established to study the injury mechanism and kinematic response of occupants with different seat orientations in frontal crashes; Finally, the simulation models of active rotating seat with belt airbag were established to study the kinematic response and injury risk of occupants with rotating seat and belt airbag protection strategy. The belt airbag with a capacity of 65 L developed in this paper could effectively protect occupants in a frontal collision at an inflation time of 200 ms and a total inflation mass of 31.4 g; The occupant safety strategy based on rotating seat and belt airbag can effectively reduce the risk of collision injury of occupants with different seat orientations in autonomous vehicle.

Predicting human chronological age via AI analysis of dorsal hand versus facial images: A study in a cohort of Indian females.

Predicting a person's chronological age (CA) from visible skin features using artificial intelligence (AI) is now commonplace. Often, convolutional neural network (CNN) models are built using images of the face as biometric data. However, hands hold telltale signs of a person's age. To determine the utility of using only hand images in predicting CA, we developed two deep CNNs based on 1) dorsal hand images (H) and 2) frontal face images (F). Subjects (n = 1454) were Indian women, 20-80 years, across three geographic cohorts (Mumbai, New Delhi and Bangalore) and having a broad variation in skin tones. Images were randomised: 70% of F and 70% of H were used to train CNNs. The remaining 30% of F and H were retained for validation. CNN validation showed mean absolute error for predicting CA using F and H of 4.1 and 4.7 years, respectively. In both cases correlations of predicted and actual age were statistically significant (r(F) = 0.93, r(H) = 0.90). The CNNs for F and H were validated for dark and light skin tones. Finally, by blurring or accentuating visible features on specific regions of the hand and face, we identified those features that contributed to the CNN models. For the face, areas of the inner eye corner and around the mouth were most important for age prediction. For the hands, knuckle texture was a key driver for age prediction. Collectively, for AI estimates of CA, CNNs based solely on hand images are a viable alternative and comparable to CNNs based on facial images.

Deep learning-based skin care product recommendation: A focus on cosmetic ingredient analysis and facial skin conditions.

Recommendations for cosmetics are gaining popularity, but they are not being made with consideration of the analysis of cosmetic ingredients, which customers consider important when selecting cosmetics. This article aims to propose a method for estimating the efficacy of cosmetics based on their ingredients and introduces a system that recommends personalized products for consumers, combined with AI skin analysis. We constructed a deep neural network architecture to analyze sequentially arranged cosmetic ingredients in the product and incorporated skin analysis models to get the precise skin status of users from frontal face images. Our recommendation system makes decisions based on the results optimized for the individual. Our cosmetic recommendation system has shown its effectiveness through reliable evaluation metrics, and numerous examples have demonstrated its ability to make reasonable recommendations for various skin problems. The result shows that deep learning methods can be used to predict the effects of products based on their cosmetic ingredients and are available for use in personalized cosmetic recommendations.

Anthropometric analysis of the faces of Chinese beauty pageant winners, from the early 20th century to the present.

The aim of this study was to investigate the aesthetic characteristics of Chinese beauty through an anthropometric analysis of beauty pageant winners from 1952 to 2021. Photographs from 5 renowned Chinese beauty pageants (Miss World Pageant, Miss Earth Pageant, Miss Universe Pageant, Miss International Pageant, and Miss Chinese Cosmos Pageant) were gathered and a total of 294 frontal and 31 lateral photographs were collected. The photos were divided into 3 time phases: phase I (1952-1977): from the first time when Chinese girls participated in beauty pageants, up until the time of reform and opening; phase II (1978-2002): from reform and opening to the re-opening of the Miss Chinese Cosmos Pageant; phase III (2003-2021): to the present. The horizontal corneal diameter (11.52mm) was set as the reference value for the frontal face. Among the 35 anthropometric items of frontal faces (26 distances and 9 angles) and 41 anthropometric items of lateral faces (33 distances and 8 angles), 21 anthropometric items (16 distances and 5 angles) and 8 proportional indices of frontal faces showed significant differences. On lateral pictures, 2 angles showed significant differences among the 3 phases. Regarding the anthropometric measurements that decreased across the 3 phases, the width of the face, midface, and lower face exhibited the greatest changes. The lower face width showed the most striking changes, with a decrease from 128.5 to 112.0mm. The quadratic regression revealed a significant turning point after 1978, coinciding with the period of Chinese reform and opening, which had a profound impact on aesthetics.