Ongoing

Touchless Palmprint Recognition Using Locally Attentive Convolutional Neural Networks

In this work, we try to propose a novel approach to touchless palmprint recognition using locally attentive CNNs. Its a CNN architecture with an image and its sub patches as input. With this methodology, we aim to reduce the inaccuracy of prediction of traditional methods by combining local and global image information into a single model. We use bilinear pooling to combine the feature descriptors of the local and global CNNs and use a inner product regularization while training.

*Paper work is ongoing

Eyeblink Detection using an Ensemble Predictor and Z-score Thresholding

This work aims at detection of eyeblinks in low exposure and complex environments. We propose multiple set of features which can be used to detect eyeblinks using a z-score based peak detection algorithm.

*Paper work is ongoing

Publications

International Journal of Information Systems & Management Science, Vol. 1, No. 1, 2018

Depression is a medical illness that affects an individual negatively by changing the way one feels, think and act but luckily, it's treatable, so the problem is its detection which we can solve with the help of machine learning. A model is developed to detect whether a person is suffering from depression or not using the prosodic features (pitch, tone, rhythm) of their voice which are promising indicators of depression. The model is trained by the data provided by the DAIC-WOZ that contains clinical interviews taken by the virtual interviewer called Ellie, which is then preprocessed using sox and system programming to remove the voice of virtual interviewer. Feature extraction is done by making spectrograms of each audio file using Librosa library. Then these spectrograms are passed into a convolution neural network with average pooling layers, dropout, He-initialisation, batch normalization, exponential linear unit activation function, and Nesterov accelerated gradients optimizer.It has been observed from the experiments that the prediction is made with an average F1 score of 0.93.

Optimizing Fast Fourier Transform (FFT) Image Compression using Intelligent Water Drop (IWD) Algorithm

Multimedia Tools and Applications, Springer

A new technique for optimizing image compression is proposed using the Fast Fourier Transform (FFT) and Intelligent Water Drop (IWD) algorithm. Specific methods are needed to decrease the number of bits required to represent digital images efficiently. This technique of digital image processing is called image compression. A wide range of techniques have been developed over the years, and novel approaches continue to emerge. IWD-based FFT Compression is a new method, and we expect compression findings to be much better than the current method. The work aims to enhance the degree of compression of the image while maintaining the features that contribute most. It optimizes the threshold values using swarm-based optimization technique (IWD) and compares the results in terms of Structural Similarity Index (SSIM). The criterion of structural similarity of image quality is based on the premise that the human visual system is highly adapted to obtain structural information from the scene, so a measure of structural similarity provides a reasonable estimate of the perceived image quality.

*Under Review

IEEE Sensors Letters

In this paper, we propose a novel approach to develop an intelligent automated real time surveillance system based solely on efficient tracking and facial recognition. With this architecture, we aim at compensating for the deficiencies of face recognition-based surveillance methods that is computationally intractable or unreliable to be deployed for real time automated surveillance. Our work aims to construct a standalone video surveillance system which is capable of detecting, identifying and tracking persons in a video feed, snapshotting interesting physical feature changes and transmitting this information along with the snapshots and identification parameter to the user. Specifically, we adopt a computationally cheap and accurate object tracker Deep Sort and combine it with a very accurate and computationally heavy facial recognition model FaceNet. Our method shows a 115% increase in runtime for the aforementioned combined architecture. The system also consists of a smartphone interface for efficient searching and indexing of identifications across timestamps.

Lecture Notes in Networks and Systems, Springer

Computer Vision is the way in which the computer perceives a certain image. Background and Foreground detection of an image is based on the concept of computer vision. Traditional approaches in background and foreground detection of an image imply clustering algorithms like K-Means Clustering, Gaussian Mixture Model to compute the result in the spatial domain. In spatial domain, we take into account the pixels of an image to classify them as background or foreground. In this paper, we have reviewed the process already been done in spatial domain, then we study some of the state-of-art background detection techniques in digital image processing and propose a novel approach to shift the domain of input images to frequency while applying the same algorithms for detection. We have used the concept in which we take into consideration the frequency rather than pixels of an image. The concept of Fast Fourier Transform (FFT) has been used to determine the frequency. With this solution concept, we aim at reducing the variance in input image by smoothing out the frequency domain image and experimentally demonstrate that the transition into the frequency domain outperforms the majority of techniques employed in spatial domain for background detection.

*Accepted for publication

Soft Computing, Springer

This paper proposes a methodology to develop a reliable and computationally inexpensive real time automatic accident detection system that can be deployed with minimum hardware requirements. We also propose Mini-YOLO, a distilled version of YOLOv3 with comparable accuracy and very low computational complexity.

*Accepted for publication

Real Time monitoring and detecting turf soil temperature using IOT devices and ARIMA

Microprocessors and Microsystems, Elsevier

In this work we have proposed a method to predict monthly turf soil temperature using the previous data and collect the relevant parameters using iot sensors such as temperature, humidity, no special geographic information was required to do the prediction. A turf extends for a width of about 0.61m above soil. This method requires no special hardware and uses the popular time series analysis employing ARIMA for the nonlinear data. We have taken 2 separate regions in North Dakota and performed our analysis over a period of 20 years (2000 -2019).

*Accepted for publication