kourosh kiani

Given the increasing daily volume of videos generated by security cameras in personal and public spaces, monitoring the activities present in videos has become crucial. Many video surveillance systems are designed to verify performance accuracy and provide alerts during the occurrence of abnormal activities. In this regard, various intelligent models have been proposed for detecting activities in videos. Considering recent advances in artificial intelligence, particularly deep learning, this paper introduces a model based on the Transformer network. To reduce computational complexity, keypoints of the human body are utilized in this approach. Fifteen key body points are input into the Transformer model, leveraging parallel processing during training and a self-attention mechanism. This enhances the speed and accuracy of the model. Experimental results on the JHMDB public database indicate an improvement in the accuracy of detecting abnormal activities compared to baseline models. Keywords: Video processing, Video surveillance, Abnormal activities, Deep learning, Transformer Network.

Fingerprint verification has emerged as a cornerstone of personal identity authentication. This research introduces a deep learning-based framework for enhancing the accuracy of this critical process. By integrating a pre-trained Inception model with a custom-designed architecture, we propose a model that effectively extracts discriminative features from fingerprint images. To this end, the input fingerprint image is aligned to a base fingerprint through minutiae vector comparison. The aligned input fingerprint is then subtracted from the base fingerprint to generate a residual image. This residual image, along with the aligned input fingerprint and the base fingerprint, constitutes the three input channels for a pre-trained Inception model. Our main contribution lies in the alignment of fingerprint minutiae, followed by the construction of a color fingerprint representation. Moreover, we collected a dataset, including 200 fingerprint images corresponding to 20 persons, for fingerprint verification. The proposed method is evaluated on two distinct datasets, demonstrating its superiority over existing state-of-the-art techniques. With a verification accuracy of 99.40% on the public Hong Kong Dataset, our approach establishes a new benchmark in fingerprint verification. This research holds the potential for applications in various domains, including law enforcement, border control, and secure access systems.

The classification of emotions using electroencephalography (EEG) signals is inherently challenging due to the intricate nature of brain activity. Overcoming inconsistencies in EEG signals and establishing a universally applicable sentiment analysis model are essential objectives. This study introduces an innovative approach to cross-subject emotion recognition, employing a genetic algorithm (GA) to eliminate non-informative frames. Then, the optimal frames identified by the GA undergo spatial feature extraction using common spatial patterns (CSP) and the logarithm of variance. Subsequently, these features are input into a Transformer network to capture spatial-temporal features, and the emotion classification is executed using a fully connected (FC) layer with a Softmax activation function. Therefore, the innovations of this paper include using a limited number of channels for emotion classification without sacrificing accuracy, selecting optimal signal segments using the GA, and employing the Transformer network for high-accuracy and high-speed classification. The proposed method undergoes evaluation on two publicly accessible datasets, SEED and SEED-V, across two distinct scenarios. Notably, it attains mean accuracy rates of 99.96% and 99.51% in the cross-subject scenario, and 99.93% and 99.43% in the multi-subject scenario for the SEED and SEED-V datasets, respectively. Noteworthy is the outperformance of the proposed method over the state-of-the-art (SOTA) in both scenarios for both datasets, thus underscoring its superior efficacy. Additionally, comparing the accuracy of individual subjects with previous works in cross subject scenario further confirms the superiority of the proposed method for both datasets.

Classification of isolated digits is a fundamental challenge for many speech classification systems. Previous works on spoken digits have been limited to the numbers 0 to 9. In this paper, we propose two deep learning-based models for spoken digit recognition in the range of 0 to 599. The first model is a Convolutional Neural Network (CNN) model that uses the Mel spectrogram obtained from the audio data. The second model uses the recent advances in deep sequential models, especially the Transformer model followed by a Long Short-Term Memory (LSTM) Network and a classifier. Moreover, we also collected a dataset, including audio data by a contribution of 145 people, covering the numerical range from 0 to 599. The experimental results on the collected dataset indicate a validation accuracy of 98.03%.

Gender recognition is an attractive research area in recent years. To make a user-friendly application for gender recognition, having an accurate, fast, and lightweight model applicable in a mobile device is necessary. Although successful results have been obtained using the Convolutional Neural Network (CNN), this model needs high computational resources that are not appropriate for mobile and embedded applications. To overcome this challenge and considering the recent advances in Deep Learning, in this paper, we propose a deep learning-based model for gender recognition in mobile devices using the lightweight CNN models. In this way, a pretrained CNN model, entitled Multi-Task Convolutional Neural Network (MTCNN), is used for face detection. Furthermore, the MobileFaceNet model is modified and trained using the Margin Distillation cost function. To boost the model performance, the Dense Block and Depthwise separable convolutions are used in the model. Results on six datasets confirm that the proposed model outperforms the MobileFaceNet model on six datasets with the relative accuracy improvements of 0.02%, 1.39%, 2.18%, 1.34%, 7.51%, 7.93% on the LFW, CPLFW, CFP-FP, VGG2-FP, UTKFace, and own data, respectively. In addition, we collected a dataset, including a total of 100’000 face images from both male and female in different age categories. Images of the women are with and without headgear.

Image forgery, the manipulation of an image to hide some meaningful or helpful information, is widely used to manage the large amount of information being exchanged in the form of images. There are different forms of image forgery, and copy--move forgery is the most common form of it. The copy-move forgery is easy to perform yet challenging to detect due to the similarity between the original part of the image and the copied part. In this paper, we employ a keypoint descriptor inspired by the human visual system, namely the FREAK (Fast Retina Keypoint) descriptor, for robust copy-move forgery detection. This method uses the advantages of FREAK descriptor such as fast computing and low memory load compared to SIFT, SURF, and BRISK. Finally, geometric transformation parameters are extracted and discussed. Results confirm promising results in the case of image post-processing operations such as adding noise, illumination change, and geometric transformations such as rotation and scaling.

3D mesh simplification is an important challenge in various fields. While different simplification methods have been proposed in recent years, the focus has shifted to keeping properties such as ridges and valleys along with mesh simplification. While most of the proposed models have used curvature, some challenges exist, such as the computational complexity and sensitivity to the neighborhood size. The latter can be solved by averaging several neighborhoods. This paper proposes a simple yet fast method with less sensitivity to the neighborhood size. To this end, we use the normal vector and the parameters of a probability distribution of its variations to detect the elevations, depressions (geometrical changes), and curve parts. We combine this method with the Quadric Error Metric (QEM) method to produce a hybrid method for 3D mesh simplification, preserving its elevations and depressions. Evaluation results show that our method has a lower error than the other methods.

Speeding up the system is one of the basic challenges in the real-world applications of Face Recognition (FR), whereas reducing the computational complexity can significantly increase the speed of the system. In recent years, many face recognition methods have been proposed but few of them give attention to this issue. Accordingly, in this article, we take the axis-symmetrical property of faces as a novel idea to speed up the face recognition algorithm as well as to reduce the computational complexity. Taking the axis-symmetrical property of faces leads us to use half of the face image. Proposing a face recognition system using Hidden Markov Model (HMM) as a classifier, we use the Singular Value Decomposition (SVD) to build the observation vectors. Evaluated results of the proposed system on Yale and Faces94 datasets show that the proposed system can achieve a satisfactory recognition rate with a higher speed.

Deep learning is one of the most important scopes of the Machine Learning that includes some important architectures. Deep Convolutional Neural Network is one of the attractive architectures that uses in digital image processing. In this paper, we use the Alexnet model for face recognition from input images. We fine-tune the Alexnet model by converting one or two fully connected layers to convolutional layers as well as using the suitable filters. To improve the robustness of the model in coping with the situations that some parts of the input images damaged, we use five crops of the input images including five pixel areas. Furthermore, to visualize the output of each layer, we use the Deconvolution technique in our method. The output of some convolutional and activation layers has been shown. Using this technique, we obtain the Heat-map of the image. To show the results, we use the LFW and Caltech faces datasets. After pre-processing the images of datasets, we compare the results of the Alexnet model in two states: before fine-tuning and after fine-tuning. The results show the recognition accuracy improvement of the fine-tuned models on input images.

The purpose of this study was to provide a structural model of factors affecting the elite students’ immigration tendency in Semnan University. The research method was based on a descriptive-correlational design focused on structural equations. The statistical population and sample of the research included top (male and female) students in Semnan University in the second semester of the Academic Year 2016-2017. The measurement tools included two questionnaires of Researcher-Made Questionnaire (2017) and Questionnaire on Immigration Tendency (Janalizadeh et al., 2014). The findings of this study revealed seven main factors as factors affecting the tendency to immigrate among elite students in Semnan University, among whom scientific-academic and psychological factors accounted for the highest average. The results of goodness of fit tests indicated that the final model on the factors influencing the tendency to immigrate among elite students in Semnan University was desirable. The results showed that, according to research findings, the phenomenon of immigration of elite students was not based on a single variable, rather various variables such as social, political, cultural, economic, organizational, psychological and scientific-academic factors should be simultaneously considered in connection with each other.

Present study aimed to investigate the relationship between university atmosphere and the value system with tendency to emigration among Semnan University students. The method of research was descriptive-correlational. The study population consisted of university students of Semnan (13089 students). Using random stratified sampling and Morgan table, 384 of them were selected as sample. Measurement tools included questionnaire of tendency to emigration (Janalizadeh et al, 2014), questionnaire of university atmosphere (Nosrati et al, 2015), and questionnaire of Value Survey (Schwartz, 2004). Reliability was calculated with Cronbach alpha and was respectively (82.0), (87/0) and (91/0) for each instrument. Based on research findings the tendency of migration among students was ranging from moderate to high. Also, all dimensions of the university atmosphere and value system showed relationship with and were able to forecast students’’ tendency to migrate.