*Last updated: January 2019
[Current]
[Details will be added soon] PATRON: Perpetual Abnormality Tracking with Robust Ochlocracy Neutralization
[Recent Works]
[paper] [poster] Machine Cognition of Violence in Videos using Novel Outlier-Resistant VLAD. Understanding highly accurate and real-time violent actions from surveillance videos is a demanding challenge. Our primary contribution of this work is divided into two parts. Firstly, we propose a computationally efficient Bag-of-Words (BoW) pipeline along with improved accuracy of violent videos classification. The novel pipeline’s feature extraction stage is implemented with densely sampled Histogram of Oriented Gradients (HOG) and Histogram of Optical Flow (HOF) descriptors rather than Space-Time Interest Point (STIP) based extraction. Secondly, in encoding stage, we propose Outlier-Resistant VLAD (OR-VLAD), a novel higher order statistics-based feature encoding, to improve the original VLAD performance. In classification, efficient Linear Support Vector Machine (LSVM) is employed. The performance of the proposed pipeline is evaluated with three popular violent action datasets. On comparison, our pipeline achieved near perfect classification accuracies over three standard video datasets, outperforming most state-of-the-art approaches and having very low number of vocabulary size compared to previous BoW Models.
[paper] An algorithmic approach to estimate cognitive aesthetics of images relative to ground truth of human psychology through a large user study. This research introduces a learning model that estimates the cognitive perception of aesthetics. Taking psychology into account, this bridges the gap between human and machine. The goal is to build a machine-learning model that can estimate beauty in images perceived by human eyes. We have summand our research [Firoze, A., Osman, T., Psyche, S. S., & Rahman, R. M. (2018). Scoring photographic rule of thirds in a large MIRFLICKR dataset: A showdown between machine perception and human perception of image aesthetics. Asian Conference on Intelligent Information and Database Systems (pp. 466–475), Springer; Osman, T., Psyche, S. S., Deb, T., Firoze, A., & Rahman, R. M. (2018). Differential color harmony: A robust approach for extracting Harmonic Color features and perceive aesthetics in a large dataset. International Conference on Big Data and Cloud Computing, Springer] together with the idea of humans’ personal preferences and achieved higher than state of the art performances. An extensive user study (374 participants) has been conducted to support claims. Several photographical compositional metrics have been used. Colour gradient, rule of thirds and human subject’s psychology has been picked as features. The consideration of user’s perspective or psychology is one of the key contributions of this research.
[paper] Differential Color Harmony: A Robust Approach for Extracting Harmonic Color Features and Perceive Aesthetics in a Large Dataset. This research work introduces and describes a robust method for extracting harmonic color features (HCF) and verifies its validity by predicting visual aesthetics of a large dataset against a large human survey. This work is a continuation of our previous research [1] where we demonstrated machine’s capability of understanding aesthetics with respect to the rule of thirds [2]. In this research, we have successfully devised a method of extracting HCF and trained a model that can perceive beauty with a root mean squared error of 1.115 +/- 0.196 on a scale of 0 to 5. In contrast to classic segmentation approaches, we have used HSV color scale gradients and differentials to extract the HCFs. Due to reduced computations, Differential Color Harmony is quite suitable for big data and fast computing. We have used a large dataset of 5000 images from the standard MIRFLICKR [12] dataset and conducted a survey where participants measured the beauty of these images. We have used these data to train classifiers and regression models and, verified our approach by making the machine perceive beauty against human perceived beauty.
[paper] Deep Learning and Data Balancing Approaches in Mining Hospital Surveillance Data. A number of classifier models on hospital surveillance data to classify admitted patients according to their critical conditions with an emphasis to deep learning paradigms, namely convolutional neural network, were used in this research. Three class labels were used to distinguish the criticality of the admitted 25,261 patients. The authors have set forth two distinct approaches to address the unbalance nature of data. They used multilayer perceptron (MLP), convolutional neural network (CNN), and multinomial logistic regression classifications and finally compared the performance of our models with the models developed by Firoze, Hasan and Rahman (2013). After comparison, the authors show that one of the models, including convolutional neural network based on deep learning, surpasses most models in terms of classification performance in contingent with training times and epochs. The trade-off is computational power for which—to achieve optimal accuracy—multiple CUDA cores are necessary. The authors achieved stable improvement of classification for their model using CNN.
[paper] [presentation (includes audio)] Face Recognition Time Reduction Based on Partitioned Faces without Compromising Accuracy and a Review of state-of-the-art Face Recognition Approaches. In this paper, the main objective is to make face recognition system faster by reducing recognition time without compromising accuracy for a constrained environment i.e. classroom, and provide a comparative review of state-of-the-art and classical approaches considering multiple faces that are at variable distance from the camera in the same image. This makes it a more challenging problem. Several models have been developed to partition the faces from a test image into three different levels. We have developed model hybridization by applying some classical but faster face recognition models namely Eigenfaces, Fisherfaces, Local Binary Patterns (LBP), and state-of-the-art yet relatively slower Convolutional Neural Network Model (CNN). Our proposed model hybridization technique based on different levels done by face partitioning has achieved approximately 33.43% faster performance than CNN while maintaining accuracy same as of CNN of our own dataset of faces of a classroom of 15 students while a class was going. The faces were of different students in different places, positions, poses and lighting. The objective of our research is not to enumerate and show how large a dataset we can identify by face, rather it is more interesting. We are interested in recognizing multiple faces at different distances from camera (hence, varying size, posture etc.) which calls for a unique approach as opposed to large dataset headshots from different angles of single people. We used different classification models for different levels of distance from the camera to achieve this faster response, making it a novel hybrid model.