Enhancing the Accuracies of Age Estimation With Heterogeneous Databases Using Modified CycleGAN

Multi-model classification method in heterogeneous image databases

Age estimation using facial images is applicable in various fields, such as age-targeted marketing, analysis of demand and preference for goods, skin care, remote medical service, and age statistics, for describing a specific place. However, if a low-resolution camera is used to capture the images, or facial images are obtained from the subjects standing afar, the resolution of the images is degraded. In such a case, information regarding wrinkles and the texture of the face are lost, and features that are crucial for age estimation cannot be obtained. Existing studies on age estimation did not consider the degradation of resolution but used only high-resolution facial images. To overcome this limitation, this paper proposes a deep convolutional neural network (CNN)-based age estimation method that reconstructs low-resolution facial images as high-resolution images using a conditional generative adversarial network (GAN), and then uses the images as inputs. An experiment is conducted using two open databases (PAL and MORPH databases). The results demonstrate that the proposed method achieves higher accuracy in high-resolution reconstruction and age estimation than the state-of-the art methods.

We are researching about an age and gender estimation based on wrinkle texture and color of facial images. A preliminary questionnaire (enquete) examination on how effectively the facial images could be used for gender and age estimations was executed by using 300 different faces and 21 examinees. Wrinkles appearing in the face and the shape and size of the facial parts are selected to model the age and gender estimation in this research basing on this enquete. Basing on this preliminary consideration, an image-processing algorithm for wrinkle modeling, was proposed. In addition, a method for making relationships between facial images and their keywords was proposed by using the latent semantic indexing. An efficient interface for displaying the relationships among keywords and facial images has been introduced with 2D/3D-feature space defined by the indexing.

Biometric technology is considered as more effective technology for identification and recognition of individuals in most secure environments. Face recognition is a technique based on physiological biometric characteristics. In this paper, we propose Face Recognition using DT-CWT and LBP Features (FRDL) for different databases. The original face image is resized to uniform dimensions of 2 <sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">p</sup> × 2 <sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">q</sup> . The five levels Dual-Tree Complex Wavelet Transform (DT-CWT) is applied on face image to obtain DT-CWT coefficients. The matrix of DT-CWT coefficients is segmented in to 3×3 matrixes. The Local Binary Pattern (LBP) algorithm is applied on each 3×3 matrix to get final features. The Euclidean Distance (ED) is used to compare features of test face image with data base images. It is observed that the values of False Rejection Rate (FRR), False Acceptance Rate (FAR) and Total Success Rate (TSR) are better in the proposed model compare to existing method.

We are researching about an age and gender estimation based on wrinkle texture and color of facial images. Preliminary questionnaire (enquete) examination of how effectively the facial images could be used for gender and age estimations was executed by using 300 different faces and 21 examinees. Wrinkles appeared in the face and the shape and size of the facial parts are selected to model the age and gender estimation in this research based on this enquete. Based on this preliminary consideration, an image processing algorithm for wrinkle modeling was proposed in this paper. In addition, a method for making relationships between facial images and their keywords was proposed by using the latent semantic indexing. An efficient interface for displaying the relationships among keywords and facial images has been introduced with 2D/3D-feature space defined by the indexing.

Age Specific Human-Computer Interaction (ASHCI) has vast potential applications in daily life. However, automatic age estimation technique is still underdeveloped. One of the main reasons is that the aging effects on human faces present several unique characteristics which make age estimation a challenging task that requires non-standard classification approaches. According to the speciality of the facial aging effects, this paper proposes the AGES (AGing pattErn Subspace) method for automatic age estimation. The basic idea is to model the aging pattern, which is defined as a sequence of personal aging face images, by learning a representative subspace. The proper aging pattern for an unseen face image is then determined by the projection in the subspace that can best reconstruct the face image, while the position of the face image in that aging pattern will indicate its age. The AGES method has shown encouraging performance in the comparative experiments either as an age estimator or as an age range estimator.

Many age estimation methods have been proposed for various applications such as Age Specific Human Computer Interaction (ASHCI) system, age simulation system and so on. Because the performance of the age estimation is greatly affected by the aging feature, the aging feature extraction from facial images is very important. The aging features used in previous works can be divided into global and local features. As global features, Active Appearance Models (AAM) was mainly used for age estimation in previous works. However, AAM is not enough to represent local features such as wrinkle and skin. Therefore, the research about local features is required. In previous works, local features were generally used to determine age group rather than detailed age, and the comparative studies about various local features extraction methods were not conducted. In this paper, the performances of sobel filter, difference image between original and smoothed image, ideal high pass filter (IHPF), gaussian high pass filter (GHPF), Haar and Daubechies discrete wavelet transform (DWT) are compared for extracting local features and detailed age estimation is performed by Support Vector Regression (SVR) on BERC and PAL aging database. The experimental results show that local features can be used for detailed age estimation and GHPF gives a better performance than other methods.

The datasets used in recent age estimation studies largely consist of two races (i.e., Asians or Westerners), and despite the large amount of data available, the problems regarding age–class imbalances still arise, owing to different age distributions. This causes overfitting in training process, reducing the generality of the age estimation. This problem typically occurs in homogeneous datasets, e.g., using the same Asian database for training and testing or using a database with the same age range for training and testing. Consequently, the problems arise in heterogeneous datasets, e.g., using an Asian database in training and a Westerner database in testing or using databases of different age ranges in training and testing, and the accuracy inevitably degrades when heterogeneous datasets are used for training and testing. To solve these problems, we proposes an enhanced cycle generative adversarial network (CycleGAN)-based heterogeneous race and age image transformation technique, which can transform the images of one race and age range to those of different race and age range. The encoder and decoder of the generator in the proposed enhanced CycleGAN include residual connections, thereby preventing information loss as much as possible as the layer deepens. In addition, the generator of the enhanced CycleGAN uses identity loss and age loss functions between the generator-produced image and a multi-channel input image obtained through 3D one-hot encoding. Through this, the training is directed to increasing the similarity not only between the images but also between the age class labels. And the enhanced CycleGAN uses a second discriminator in addition to the existing discriminator, thereby addressing a problem in which training is not properly performed when the discriminator converges too fast relative to the generator in a conventional CycleGAN. Experiments with three open databases demonstrated that our method outperforms state-of-the-art methods for facial image transformation and age estimation.

Changes in illumination condition, pose, facial expression and others is not an easy task in face recognition. Solving these problems requires not only a feature extraction method that can generate distinct features for each class of image but requires other additional technique that able to improve the overall classification accuracy. This paper presents the face recognition using combined features of original and pre-processed face images. This technique is experimented using orthogonal moments namely Zernike moments (ZMs) and Krawtchouk moments (KMs). The classification technique used in the recognition stage is Euclidean square distance or Nearest Neighbour (NN) classifier. Database face images from Olivetti research laboratory (ORL) consisting of 40 subjects of 10 images each where none of them are identical, is used in the experiments. The face images vary in position, rotation, scale and expression, with and without spectacles. From the experiments, the new technique is able to improve the classification accuracy significantly.

In this paper, we present a Face Recognition method based on the combination of the Gabor Wavelets (GW) and the Phase Congruency (PC) method [1]. While the GW method allows an efficient space-frequency analysis to code the Face Image points, it requires significant down-sampling, in order to end up with a reasonably small-size feature vector. Now, the Phase Congruency method can be used as an efficient features detector, to uncover the salient local features of the face and can be computed using the GW Transform. An additional advantage of the Phase Congruency method is its insensitivity to variations in image illumination and contrast. The method proposed is to use the PC method in determining the salient points of the Face Image, thus replacing the random down-sampling method. The first step in our proposed method is to obtain the local frequency information of a face image using a multi-scale, multi-orientation set of Gabor filters, from which the Phase Congruency image is computed. We then use this method to select a limited number of maximum-value points, whose values are taken from the GW- transformed face images, forming a feature vector of about 5000 components. Preliminary results using the ORL face dataset show 98% recognition rate even without the use of the PCA method. Upon using the PCA method we can further reduce the number of components to 45 while still retaining the recognition rate of 98%. This as compared to a recognition rate of 96% using GW and PCA with a random 64:1 down-sampling on the GW coefficients. The proposed combined method, allows in addition to an efficient reduction in the number of image points used, the derivation of both the PC image and the GW feature vector representation, using a single GW transform.

Age estimation from facial images based on Gabor feature fusion and the CIASO‐SA algorithm

In this paper, we propose age and gender estimation system by vairous features. Age and gender has a lot of characteristics. These characteristics are one of the difficult cognitive process in human interaction. If we can extract the important feature of this cognitive process, it is considered that the age and gender estimation by the machine becomes possible. Therefore, we propose a method of age and gender feature extraction and estimation using the face image. In this paper, the age of the face means apparent-age that is based on the human perception of age. Moreover, person's aging and gender difference appear in the faces. For example, the pigmented spot, the wrinkle, sagging skin, shape, color of skin, and so on. Thus, we extract these several features for age and gender estimation. Furthermore, we estimate a continuous age and gender using a neural network (NN).

Background and PurposeSleepiness is a major factor in accidents or errors. Even in these days of promoting system automation, users should monitor the state of the autonomous system or take over in the event of an emergency. Thus, it is important for systems to understand user’s arousal level and manage user’s arousal level appropriately. In recent years, a method for maintaining arousal levels has been proposed by using artificial intelligent agents to converse with the user. The interaction method used by such a system is considered to be a more natural way to maintain arousal level than the conventional method of using stimulation such as a beep sound. Therefore, we anticipate realizing a system that not only detects sleepiness and responds to it reactively, but also predicts a decreasing arousal level in advance and responds to it proactively. In this study, aiming at realizing such a system, we attempt to construct a model to predict decreasing arousal levels in advance. We think this will lead to the optimization of system interaction. In the assessment of arousal level, there is a method to assess a user’s arousal level in five levels based on facial expression by trained raters. Therefore, we assumed to construct a model predicting the five stages of arousal level in the method based on features of face images. MethodIn the experiment to obtain data for model construction, autonomous driving was assumed as an example of an autonomous system. Participants monitored autonomous driving for an hour. Participants could only change lanes by pressing a button, and all other operations were handled by the autonomous system. During the task, participants responded to the arousal level in five levels by pressing a button every 30 seconds. In addition, participant’s face images were recorded at 60 Hz. Three male and three female participants took part in the experiments, and each participant completed the task three times. From face images, we used features of a texture distribution by Local Binary Patterns Histogram (LBPH) and features obtained by embedding using FaceNet for model construction. Using these features and subjective assessment of arousal level, we constructed a model to predict future arousal level in five levels by machine learning. ResultsThe bias of the output value was less when using a neural network than when using random forest or support vector machine. The accuracies of prediction arousal level after 30 seconds from the current features are around 35% for learning individual data. Although the accuracies decrease as the target time for prediction became far away from the present time, the results up to 120 seconds showed that the accuracy was greater than 30% and the rate of the case where the deviation between predicted and actual values is less than one was greater than 60%, which means that it is possible to suppress the cases where large deviations occur. We think that the model is expected to be used in a proactive system that predicts and responds to decreases in arousal levels in advance.

In human age estimation, an exact age or age group is assigned automatically to a given facial image. Due to changes in facial features caused by wrinkles and muscle sag with the increase in age, age estimation becomes difficult and complex. Variability in facial expressions, illumination conditions, environment, pose, plastic surgery, occlusion, image quality, and facial makeup gives rise to some of the challenges that affect the accuracy of age estimation. With the newly developing real-time and age-related applications, there is an increasing demand for more accurate estimation of age. Several methodologies have been developed by researchers to handle this requirement. The work reported here represents an exhaustive survey on face age estimation. Discussed here are five major models of age image representation and various age learning algorithms, a review of the existing aging databases has been made, and the latest state-of-the-art methods for the estimation of age are summarized.

Methods for age estimation in children involve measurements of: (1) the projections of open apices and tooth heights (T), (2) the total area of the carpal bones and the epiphyses of the ulna and radius (HW), and (3) a combination of these parameters (THW). This study aimed to compare the accuracy of the original formulae for T, HW, and THW in Black South African (BSA) and White South African (WSA) samples of children and adolescents. The sample comprised 556 subjects, 164 BSA (71 males and 93 females) and 392 WSA (171 males and 221 females) aged 6 to 16 years. The main sample comprised 556 subjects and was divided into two parts. The first part was used to develop new formulae specific to the South African population, while the second part, consisting of 183 subjects and referred to as the test sample, was used to evaluate the accuracy of these formulae and to quantify their uncertainty when applied to real cases. Adjusted R2 and standard errors of the estimate (SEEs) were calculated for each regression model. The mean absolute difference or error (MAE) between estimated and chronological age and the percentage of individuals within specific age ranges of ± 0.25, ± 0.5, ± 1, and ± 2 years of chronological age was used to compare the accuracy of formulae. Teeth showed the most significant underestimation in the training sample, -0.49 years in males and - 0.29 years in females, as much as -0.70 years in WSA males, while BSA males were overestimated by + 0.03 years. Combining teeth and hand-wrist, the new formula showed the best performance in age estimation compared to only teeth or hand-wrist, except in WSA males (-0.51 years). New South African-specific formula improved accuracy, especially in males, suggesting their suitability alongside the original Cameriere et al. methods for precise age estimation. The original regression formulae by Cameriere et al., along with the newly developed formulaespecific to South African populations, shows promise for age estimation, considering potential estimation errors and the accuracy rates demonstrated in this study.