Joseph H. R. Isaac

Bio: Joseph H. R. Isaac is an academic researcher from Indian Institute of Technology Madras. The author has contributed to research in topics: Pose & Multispectral pattern recognition. The author has an hindex of 2, co-authored 7 publications receiving 32 citations. Previous affiliations of Joseph H. R. Isaac include Jeppiaar Engineering College.

Urban landcover classification from multispectral image data using optimized AdaBoosted random forests

Abstract: With an ever growing need to classify multispectral images, the accuracy of the classification becomes a matter of concern, especially when mapping heterogeneous environments such as urban areas. N...

Novel virtual reality based training system for fine motor skills: Towards developing a robotic surgery training system.

Abstract: BACKGROUND Training surgeons to use surgical robots are becoming part of surgical training curricula. We propose a novel method of training fine-motor skills such as Microscopic Selection Task (MST) for robot-assisted surgery using virtual reality (VR) with objective quantification of performance. We also introduce vibrotactile feedback (VTFB) to study its impact on training performance. METHODS We use a VR-based environment to perform MST with varying degrees of difficulties. Using a well-known human-computer interaction paradigm and incorporating VTFB, we quantify the performance: speed, precision and accuracy. RESULTS MST with VTFB showed statistically significant improvement in performance metrics leading to faster completion of MST with higher precision and accuracy compared to that without VTFB. DISCUSSION The addition of VTFB to VR-based training for robot-assisted surgeries may improve performance outcomes in real robotic surgery. VTFB, along with proposed performance metrics, can be used in training curricula for robot-assisted surgeries.

Reverse Circle Cipher for personal and network security

Abstract: Many data encryption techniques have been employed to ensure both personal data security and network security. But few have been successful in merging both under one roof. The block cipher techniques commonly used for personal security such as DES and AES run multiple passes over each block making them ineffective for real time data transfer. Also, ciphers for network security such as Diffie-Hellman and RSA require large number of bits. This paper suggests a simple block cipher scheme to effectively reduce both time and space complexities and still provide adequate security for both security domains. The proposed Reverse Circle Cipher uses ‘circular substitution’ and ‘reversal transposition’ to exploit the benefits of both confusion and diffusion. This scheme uses an arbitrarily variable key length which may even be equal to the length of the plaintext or as small as a few bits coupled with an arbitrary reversal factor. This method of encryption can be utilized within stand alone systems for personal data security or even streamed into real time packet transfer for network security. This paper also analyses the effectiveness of the algorithm with respect to the size of the plaintext and frequency distribution within the ciphertext.

Passive Probing Perception: Effect of Latency in Visual-Haptic Feedback

Abstract: Latency is detrimental to haptic systems, specifically in networked telepresence systems. Although the latency effect on stiffness is well studied in the literature, it is not very clear if the latency effects on the stiffness perception are due to the displacement perception or the force perception. In this study, we propose passive probing which involves force perception alone, without any displacement of the finger, for studying latency effects. A psychophysical experiment is conducted with a set of artificially induced latencies which provides a quantitative measure of the effect of these latencies on three parameters: Just Noticeable Difference (JND), the time taken to reach the reference forces and the maximum overshoot. The results showed that the latency has a significant effect on the user ability in task performance after 185.5 ms. From the observation, the latency effect on JND in passive probing is similar to that of the stiffness perception (active probing) task which shows that the effect is significantly higher on the force perception rather than on the displacement perception.

Effect of Control Movement Scale on Visual Haptic Interactions

Abstract: Although the human hand is a complex system which can perform multiple actions, when the kinaesthetic actions are scaled in a system, the applications are limitless. In this paper, we examine the effect of control movement scale on user’s kinaesthetic actions. We use the Fitts’ Law for quantifying the user’s performance on different scales and to verify if higher control movement scale, in general, can be better than natural movements in tasks which require extended accuracy. The experiment consists of a Wacom™ tablet as an input device connected to a system. The tablet provides means for scaling the kinaesthetic input movement of a user. The experiment is a modified version of the classical multi-directional tapping task. It was performed on 16 healthy participants with ages between 20 to 48 years. The Fitts’ regressions were visualised and the Z-scores were computed. It was found that the performance of the participants increases with the scale and has an optimum scale at 1:3.3 before reducing rapidly. Future works include experiments involving 3D models and other haptic input devices.

Classification of stroke disease using machine learning algorithms

Abstract: This paper presents a prototype to classify stroke that combines text mining tools and machine learning algorithms. Machine learning can be portrayed as a significant tracker in areas like surveillance, medicine, data management with the aid of suitably trained machine learning algorithms. Data mining techniques applied in this work give an overall review about the tracking of information with respect to semantic as well as syntactic perspectives. The proposed idea is to mine patients’ symptoms from the case sheets and train the system with the acquired data. In the data collection phase, the case sheets of 507 patients were collected from Sugam Multispecialty Hospital, Kumbakonam, Tamil Nadu, India. Next, the case sheets were mined using tagging and maximum entropy methodologies, and the proposed stemmer extracts the common and unique set of attributes to classify the strokes. Then, the processed data were fed into various machine learning algorithms such as artificial neural networks, support vector machine, boosting and bagging and random forests. Among these algorithms, artificial neural networks trained with a stochastic gradient descent algorithm outperformed the other algorithms with a higher classification accuracy of 95% and a smaller standard deviation of 14.69.

Improving Land Use/Cover Classification with a Multiple Classifier System Using AdaBoost Integration Technique

Abstract: Guangzhou has experienced a rapid urbanization since 1978 when China initiated the economic reform, resulting in significant land use/cover changes (LUC). To produce a time series of accurate LUC dataset that can be used to study urbanization and its impacts, Landsat imagery was used to map LUC changes in Guangzhou from 1987 to 2015 at a three-year interval using a multiple classifier system (MCS). The system was based on a weighted vector to combine base classifiers of different classification algorithms, and was improved using the AdaBoost technique. The new classification method used support vector machines (SVM), C4.5 decision tree, and neural networks (ANN) as the training algorithms of the base classifiers, and produced higher overall classification accuracy (88.12%) and Kappa coefficient (0.87) than each base classifier did. The results of the experiment showed that, based on the accuracy improvement of each class, the overall accuracy was improved effectively, which combined advantages from each base classifier. The new method is of high robustness and low risk of overfitting, and is reliable and accurate, and could be used for analyzing urbanization processes and its impacts.

Multi-Temporal Land Cover Change Mapping Using Google Earth Engine and Ensemble Learning Methods

Abstract: The study deals with the application of Google Earth Engine (GEE), Landsat data and ensemble-learning methods (ELMs) to map land cover (LC) change over a decade in the Kaski district of Nepal. As Nepal has experienced extensive changes due to natural and anthropogenic activities, monitoring such changes are crucial for understanding relationships and interactions between social and natural phenomena and to promote better decision-making. The main novelty lies in applying the XGBoost classifier for LC mapping over Nepal and monitoring the decadal changes of LC using ELMs. To map the LC change, a yearly cloud-free composite Landsat image was selected for the year 2010 and 2020. Combining the annual normalized difference vegetation index, normalized difference built-up index and modified normalized difference water index, with elevation and slope data from shuttle radar topography mission, supervised classification was performed using a random forest and extreme gradient boosting ELMs. Post classification change detection, validation and accuracy assessment were executed after the preparation of the LC maps. Three evaluation indices, namely overall accuracy (OA), Kappa coefficient, and F1 score from confusion matrix reports, were calculated for all the points used for validation purposes. We have obtained an OA of 0.8792 and 0.875 for RF and 0.8926 and 0.8603 for XGBoost at the 95% confidence level for 2010 and 2020 LC maps, which are better for mountainous terrain. The applied methodology could be significant in utilizing the big earth observation data and overcoming the traditional computational challenges using GEE. In addition, the quantification of changes over time would be helpful for decision-makers to understand current environmental dynamics in the study area.