i-manager's Journal on Pattern Recognition (JPR)


Volume 7 Issue 2 July - December 2020

Research Paper

Two-Pass Segmentation for Marathi Character Extraction

Madhav A. Kankhar* , C. Namrata Mahender **
*-** Department of Computer Science & Information Technology, Dr. Babasaheb Ambedkar Marathwada University, Aurangabad, Maharashtra, India.
Kankhar, M. A., and Mahender, C. N. (2020). Two-Pass Segmentation for Marathi Character Extraction. i-manager's Journal on Pattern Recognition, 7(2), 1-6. https://doi.org/10.26634/jpr.7.2.18148

Abstract

Handwriting recognition is very complex but a much needed system for various purposes. Work on English language can be seen in full swing as compared to Indian languages. Free style handwriting has various challenges due to the individual variation in writing style, along with half characters, joint characters, overlapped characters, etc. are major when considering written in Devanagari script. Computational studies of Indian languages written using Devanagari script, including Marathi, are in its primary stage. The present work focuses on proper character segmentation in Marathi words for better post-processing and recognition rate as segmentation is an important stage that affects post-processing performance. Data is collected from native speakers of the Marathi language, with 3000 samples handwritten words in Marathi language for numbers from one to ten, and this paper discusses the proposed two-pass segmentation approach to extract written Marathi script.

Research Paper

Analysing Meticulous Behavior of Learners from Non-Verbal Cues

Mageshwaran L.* , Keerthivasan S. E. **, Hariharan K.***, Rajasekaran T.****
*-**** Department of Computer Science and Engineering, SRM Valliammai Engineering College, Kattankulathur, Tamil Nadu, India.
Mageshwaran, L., Keerthivasan, S. E., Hariharan, K. and Rajasekaran, T. (2020). Analysing Meticulous Behavior of Learners from Non-Verbal Cues. i-manager's Journal on Pattern Recognition, 7(2), 7-13. https://doi.org/10.26634/jpr.7.2.18129

Abstract

Face emotion recognition is the most important part in deep learning. It is the best way to communicate in a non-verbal manner. The main issue, which has existed since the time, is that teachers are unable to determine whether or not their students have grasped the subject in class. The proposed model in this paper will address this issue. Face emotion recognition is becoming more popular with Artificial Intelligence. In our model, Convolutional Neural Networks is used for the emotion recognition. This is divided into two parts; face detection and extraction can be done using Haar cascades and the emotion will be extracted from the face using CNN on FER 2013. Then the emotion is collected continuously and the average will be stored in MySQL database. A user interface will be created to teachers and admin so that they can view the data of students at any time and also generate report on it. This proposed model will give the accuracy of 90%.

Research Paper

Robust Facial Expression Recognition based on Convolutional Neural Network in Pose and Occlusion

Shalmiya P.* , G. Thirugnanam **
*-** Department of Electronics and Communication Engineering, Government College of Technology, Coimbatore, Tamil Nadu, India.
Shalmiya, P., and Thirugnanam, G. (2020). Robust Facial Expression Recognition based on Convolutional Neural Network in Pose and Occlusion. i-manager's Journal on Pattern Recognition, 7(2), 14-22. https://doi.org/10.26634/jpr.7.2.18083

Abstract

Facial expression plays an important role in human-human communication. Facial Expression Recognition (FER) has various applications in attendance management system, service robots, intelligent tutoring system, driver fatigue monitoring, human behavior understanding, detection of mental disorders and has various medical and surveillance applications. The various limitations of FER are occlusion, illumination variation, variant pose, poor image quality, etc. This work mainly focuses on occluded and pose variations datasets. These issues have included wide interest in FER where it leads to unseen regions of input faces which brings difficulty for face identification and face recognition and also affects the feature extraction and classification process. Thus, these effects can be reduced and the expressions are recognised using Convolutional Neural Network (CNN) architecture. Finally, the performance metrics of the model has been measured with accuracy values of the system.

Research Paper

Noise Robust Speech Recognition under Noisy Environments

P. Sunitha* , V. Sailaja **, B. Vasantha Lakshmi ***
*-*** Department of Electronics and Communications Engineering, Pragati Engineering College, Andhra Pradesh, India.
Sunitha, P., Sailaja, V., and Lakshmi, B. V. (2020). Noise Robust Speech Recognition under Noisy Environments. i-manager's Journal on Pattern Recognition, 7(2), 23-28. https://doi.org/10.26634/jpr.7.2.18094

Abstract

This paper presents a new method for improving the recognition accuracy of a speech recognition system in a noisy environment by using robust speech enhancement technique with the aid of noise estimation algorithm. The robustness of a speech recognition system can be improved by improving the speech quality at signal level by means of noise suppression algorithms, feature extraction level or at modelling phase. The proposed method uses robust speech enhancement technique as a pre-processing operation to improve the recognition accuracy in presence of noise. The suggested method is evaluated in terms of recognition accuracy. The suggested method yields better results in terms of recognition accuracy in presence of eight different types of non-stationary noises under different SNR levels when compared with the baseline speech recognition system.

Research Paper

CV-SAS: Computer Vision Modeled Seamless Stress Analysis System for Workers

B. Muthu Senthil * , Dinesh E. **, Dinesh T. ***, Goutham Raj S. ****
*-**** Department of Computer Science and Engineering, SRM Valliammai Engineering College, Kattankulathur, Tamil Nadu, India.
Muthusenthil, B., Dinesh, E., Dinesh, T., and Raj, S. G. (2020). CV-SAS: Computer Vision Modeled Seamless Stress Analysis System for Workers. i-manager's Journal on Pattern Recognition, 7(2), 29-36. https://doi.org/10.26634/jpr.7.2.18127

Abstract

People spend most of their time in the workplace, often with high workloads and time pressure, a practice that contributes to increased stress levels. An accurate stress assessment method may thus be of importance to clinical intervention and diseases prevention. While different neuro imaging modalities have been proposed to detect mental stress, each modality experiences certain limitations. This paper proposes the ability of a computer to detect and analyze the stress in human beings with greater accuracy. The system uses the webcam to detect the facial features and perform a mood analysis, which is eventually used for detecting stress. Then, the features are fed into the SVM model as an input and using the machine learning approach the stress is classified.