A New Approach for Gender Classification Based on Gait Analysis

This work focuses on automatic gender and age prediction tasks from handwritten documents. This problem is of interest in a variety of fields, such as historical document analysis and forensic investigations. The challenge for automatic gender and age classification can be demonstrated by the relatively low performances of the existing methods. In addition, despite the success of CNN for gender classification, deep neural networks were never applied for age classification. The published works in this area mostly concentrate on English and Arabic languages. In addition to Arabic and English, this work also considers Hebrew, which was much less studied. Following the success of bilinear Convolutional Neural Network (B-CNN) for fine-grained classification, we propose a novel implementation of a B-CNN with ResNet blocks. To our knowledge, this is the first time the bilinear CNN is applied for writer demographics classification. In particular, this is the first attempt to apply a deep neural network for the age classification. We perform experiments on documents from three benchmark datasets written in three different languages and provide a thorough comparison with the results reported in the literature. B-ResNet was top-ranked in all tasks. In particular, B-ResNet outperformed other models on KHATT and QUWI datasets on gender classification.

Automatic gender classification from speech is an integral component of human-computer interfaces. Gender information is utilized in user authentication, speech recognizers, or human-centered intelligent agents. This study focuses on gender classification from speech spectrograms using AlexNet-inspired 2D convolutional neural networks (CNN) trained on real samples augmented with synthetic spectrograms. A generative adversarial network (GAN) is trained to produce synthetic male/female-like speech spectrograms. In limited training data experiments on LibriSpeech, augmenting a training set of 200 real samples by 800 synthetic samples reduces equal error rate of the classifier from 23.7% to 1.0%. To further test the ‘quality’ of the generated samples, in a subsequent experiment, the real training samples are progressively replaced (rather than augmented) with synthetic samples at various ratios from 0 (all original samples preserved) to 1 (all original samples replaced by synthetic ones). Depending on the system setup, substituting between 50% to 90% of the original samples with the synthetic ones is found to have a minimal impact on the classifier performance. Finally, viewing the input CNN layers as filters that select salient spectrogram features, the learned convolutional kernels and filter outputs are studied to understand which spectrogram areas receive a prominent attention in the classifier.

Gender classification based on facial features plays a central role in numerous intelligent applications such as surveillance cameras, interactive advertising, and human-computer interaction. This study aimed to compare two deep models, YOLOv9 and ResNet50, on face detection and gender classification, focusing on accuracy and inference speed. YOLOv9 performed well in terms of speed, with an inference time of 332 ms per image and a processing speed of 3 fps, and had a precision of 86.8%, 86.1% of recall, and 86.54% of F1-score. These performance characteristics make YOLOv9 suitable for real-time applications with high-speed response demands, even with moderately low classification accuracy. Conversely, ResNet50 was applied directly to gender classification after data preparation on images and had high classification accuracy, with a precision of 93.6%, 92% of recall, and 92.79% of F1-score. Its inference time was slower at 446.33 ms per image, with a 2.24 fps processing speed and a long training time of 9 hours and 18 minutes. These results show that YOLOv9 has high performance within a time scope of face detection, with reference to detecting enough faces within short timeframes with a limited number of computational resources, whereas ResNet50 has better classification accuracy. Depending on particular use case scenario demands, one corresponding model with a preferred feature can be selected: YOLOv9, if high-speed response is a concern during real-time applications, and ResNet50, if high classification accuracy is a concern.

This paper describes an analysis of body radiation frequency for the purpose of gender classification. The human radiation frequency is experimentally studied from 33 healthy human subjects of 17 males and 16 females. KNN classifier is employed for gender classification. The number of training to testing ratio was evaluated at 50 to 50, 60 to 40 and 70 to 30, to determine best classification accuracy. The data was analyzed separately of raw dataset and post-processing dataset to compare the classification results. At first, the data was classified using raw dataset and yields the classification accuracy of 93.8%. Then, the post-processing data was applied to the classifier, and it was found that the classification accuracy was improved to perfect classification on k = 5, 7, 11 and 13 to 15.

Gender classification from facial images plays a significant role in biometric technology viz. gender medicine, surveillance, electronic banking system and human computer interaction. However, it has many challenges due to variations of pose, expression, aging, race, make-up, occlusion and illumination. In the proposed system, spatially enhanced local binary pattern (SLBP) and histogram of oriented gradients (HOG) are extracted to classify the human gender with SVM classifier. This hybrid feature selection has increased the power of the proposed system due to its representation of texture micro-patterns and local shape by capturing the edge or gradient structure form the image. The gender classification accuracy is studied by using the local feature representation of the face images separately and also these features are concatenated to provide a better recognition rate. The combination of two different local descriptors provides good representation of face image and this is given to SVM classifier which classifies as male or female. Also, the proposed work is compared with other two traditional classifiers such as k-nearest neighbor and sparse representation classifier. The performance was evaluated on FERET and LFW database. The highest classification accuracy 99.1% is achieved on FERET database and 95.7% is achieved on LFW database by applying cubic SVM with fusion of SLBP and HOG features.

This paper presents a thorough study into the influence of the image resolution on automatic face gender classification. The images involved range from extremely low resolutions (2 × 1 pixels) to full face sizes (329 × 264 pixels). A comprehensive comparison of the performances achieved by two classifiers using ten different image sizes is provided by means of two performance measures Correct Classification Rate (CCR) and Geometric Mean (GMean). Single- and cross-database experiments are designed over three well-known face datasets. A detailed statistical analysis of the results revealed that a face as small as 3 × 2 pixels carries some useful information for distinguishing between genders. However, in situations where higher resolution face images are available, moderately sized faces from 22 × 18 to 90 × 72 pixels are optimal for this task. Furthermore, the performance of the classifiers was robust to the changes in the image resolution (using medium to full sizes). Only when the image resolution was reduced to 8 × 6 pixels or smaller, the classification results were significantly affected.

A query image based scene/image retrieval system is a system that analyzes the properties of a query image and identifies the class in which the image belongs and retrieves a number of images which are most alike and relevant to the query image. A scene/image classifier provides the first stage for this system. Scene classification is the process that analyzes the properties of various image features of a scene and organizes the scenes into categories. Programmed techniques for correlating scenes with labels have a high potential for improving the performance of computer vision applications such as searching, retrieval and object recognition. The scene category system uses the feature vectors extracted from image data to classify test images. We have implemented a k-nearest neighbour (kNN) classifier which uses gray scale and Scale Invariant Feature Transform (SIFT) features as input and used support vector machine (SVM) classifier that uses GIST features as input. kNN classifier is used, as it has an advantage of working with a large amount of data and SVM is robust even when the training sample has some bias. The performance measure of the classifier was evaluated using confusion matrix and classification accuracy. We considered MIT-8 scene category, leaves, Caltech-20 and deer data sets for our experiments. The classification accuracy was found to be good. Results show that classification accuracy is increased significantly for classification using GIST features in SVM classifiers. We are exploring methods to obtain optimal performance of classifiers.

In recent academic literature Sex and Gender have both become synonyms, even though distinct definitions do exist. This give rise to the question, which of those two are actually face image classifiers identifying? It will be argued and explained why CNN based classifiers will generally identify gender, while feeding face recognition feature vectors into a neural network, will tend to verify sex rather than gender. It is shown for the first time how state of the art Sex Classification can be performed using Embedded Prototype Subspace Classifiers (EPSC) and also how the projection depth can be learned efficiently. The automatic Gender classification, which is produced by the emph{InsightFace} project, is used as a baseline and compared to the results given by the EPSC, which takes the feature vectors produced by emph{InsightFace} as input. It turns out that the depth of projection needed is much larger for these face feature vectors than for an example classifying on MNIST or similar. Therefore, one important contribution is a simple method to determine the optimal depth for any kind of data. Furthermore, it is shown how the weights in the final layer can be set in order to make the choice of depth stable and independent of the kind of learning data. The resulting EPSC is extremely light weight and yet very accurate, reaching over $98\%$ accuracy for several datasets.

High accuracy remote sensed image classification technology is a long-term and continuous pursuit goal of remote sensing applications. In order to evaluate single classification algorithm accuracy, take Landsat TM image as data source, Northwest Yunnan as study area, seven types of land cover classification like Maximum Likelihood Classification has been tested, the results show that: (1)the overall classification accuracy of Maximum Likelihood Classification(MLC), Artificial Neural Network Classification(ANN), Minimum Distance Classification(MinDC) is higher, which is 82.81% and 82.26% and 66.41% respectively; the overall classification accuracy of Parallel Hexahedron Classification(Para), Spectral Information Divergence Classification(SID), Spectral Angle Classification(SAM) is low, which is 37.29%, 38.37, 53.73%, respectively. (2) from each category classification accuracy: although the overall accuracy of the Para is the lowest, it is much higher on grasslands, wetlands, forests, airport land, which is 89.59%, 94.14%, and 89.04%, respectively; the SAM, SID are good at forests classification with higher overall classification accuracy, which is 89.8% and 87.98%, respectively. Although the overall classification accuracy of ANN is very high, the classification accuracy of road, rural residential land and airport land is very low, which is 10.59%, 11% and 11.59% respectively. Other classification methods have their advantages and disadvantages. These results show that, under the same conditions, the same images with different classification methods to classify, there will be a classifier to some features has higher classification accuracy, a classifier to other objects has high classification accuracy, and therefore, we may select multi sub-classifier integration to improve the classification accuracy. © (2015) COPYRIGHT Society of Photo-Optical Instrumentation Engineers (SPIE). Downloading of the abstract is permitted for personal use only.

Gender classification is one of the most challenging problems in computer vision. Facial gender detection of neonates and children is also known as a highly demanding issue for human observers. This study proposes a novel gender classification method using frontal facial images of people. The proposed approach employs principal component analysis (PCA) and fuzzy clustering technique, respectively, for feature extraction and classification steps. In other words, PCA is applied to extract the most appropriate features from images as well as reducing the dimensionality of data. The extracted features are then used to assign the new images to appropriate classes – male or female – based on fuzzy clustering. The computational time and accuracy of the proposed method are examined together and the prominence of the proposed approach compared to most of the other well‐known competing methods is proved, especially for younger faces. Experimental results indicate the considerable classification accuracies which have been acquired for FG‐Net, Stanford and FERET databases. Meanwhile, since the proposed algorithm is relatively straightforward, its computational time is reasonable and often less than the other state‐of‐the‐art gender classification methods.

Prior work by a subset of authors led to a development of optical camera platform, which has a capability of distinguishing between human and animal movement in an outdoor environment. Once the image is classified as a human, an idea of providing additional information and insights led to the exploration of gender (men vs women) and direction (left to right vs right to left) classifications. The proposed method classifies the human gender, based on the full body image oriented in a side view manner. An additional feature is to classify the direction of the movement. In the current paper, the Convolutional Neural Networks (CNNs) are used to distinguish between men and women gender classes and to identify the direction of the movement. Furthermore, different aspects of CNN are visualized (for example, attention heat maps, t-Distributed Stochastic Neighbor Embedding plot, etc.) to provide useful insights corresponding to classifications and misclassifications. Additionally, different CNN architectures were tried to figure out the best possible choice. The classification accuracies for gender (men vs women) and direction classification (left to right vs right to left) on the test data are close to 93.3% and 94%, respectively.

In this paper, we demonstrate the need for dimensionality reduction to mitigate model overfitting on the nontrivial problem of gender classification from digital images. In this study we explore four feature selection schemes using Genetic Algorithm, Memetic Algorithms, and Random Forest, which are fed to a nonlinear support vector machine (SVM) for final classification. The performance of the model (feature) selection approaches are evaluated against two distinct datasets of facial images: FG-NET which contains toddlers to seniors and the UIUC-PAL which contains faces of adults up to seniors. This work demonstrates that feature selection can, and does, improve performance of an SVM based gender classification system significantly.

Automatic gender classification of an individual can be very useful in video-based surveillance systems and human-computer interaction systems. In this paper, we propose an approach to integrate information from multi-view gait at the feature level. First, gait energy images (GEI) are constructed from the video streams for different viewpoints. Then, the feature fusion is performed by putting GEI images and camera views together to generate a third-order tensor (x, y, view). A multilinear principal component analysis is employed to reduce dimensionality of the tensor objects which integrate all views. Compared with other methods, the proposed fusion scheme shows more effective performance for multi-view gait based gender classification.

Gender is one of the essential characteristics of personal identity but is often misused by online impostors for malicious purposes. However, men and women differ in their natural aiming movements of a hand held object in two-dimensional space due to anthropometric, biomechanical, and perceptual-motor control differences between the genders. Exploiting these natural gender differences, this paper proposes a naturalistic approach for gender classification based on mouse biometrics. Although some previous research has been done on gender classification using behavioral biometrics, most of them focuses on keystroke dynamics and, more importantly, none of them provides a comprehensive guideline for which metrics (features) of movements are actually relevant to gender classification. In this paper, we present a method for choosing metrics based on empirical evidence of natural difference in the genders. In particular, we develop a novel gender classification model and evaluate the model’s accuracy based on the data collected from a group of 94 users. Temporal, spatial, and accuracy metrics are recorded from kinematic and spatial analyses of 256 mouse movements performed by each user. A mouse signature for each user is created using least-squares regression weights determined by the influence movement target parameters (size of the target, horizontal and vertical distances moved). The efficacy of our model is validated through the use of binary logistic regressions.

A watermark is embedded into a color image. The relationship between embedding and compression process is investigated. The compression ratio (CR) is adjusted from 2 to 15, and three gray-levels of the original image are changed, their average gray-levels are individually 126, 161 and 196. From the experimental results, there is an apparent inclination about higher gray-level makes higher PSNR. There are two kinds of compression methods in the compression process, one includes color space transform and sampling, and this method has higher CR, the other has not these steps and it has lower CR. In this paper the relationship of these parameters such as the gray-level of the image, the parameters of frequency domain are also discussed and analyzed. The relation between color saturation and normalized correlation (NC) ratio has been quantitatively analyzed, it is found that the extracted watermark has the best quality when the saturation factor is set to the value of 0.2, and the NC value is lower when the processing has lower color factor. The curve of NC drops gradually when the color factor is over 0.4. The main reason of this result is also discussed in the final discussion. In our proposed approach, a systematic analysis is performed. The adaptive methods are selected to obtain the optimum conditions to embed the watermark into an image.