Amit Laddi

An augmented image gradients based supervised regression technique for iris center localization

This paper describes a robust and accurate technique for iris center localization by combining supervised regression based approach and image gradients. The proposed work consist of two stages. The first stage comprises regression approach which is based upon learning of local binary features to detect the periocular regions. In the second stage, image gradients were applied to the extracted eye patch regions to detect the accurate iris centers. The proposed augmented image gradients based supervised regression approach tested on the two publicly available challenging datasets show good accuracy. The results proved that supervised regression technique when augmented with image gradients approach improved the accuracy of iris center detection on the face image acquired under unconstraint conditions. The outcome of the proposed work suggests that by augmenting effective unsupervised techniques such as image gradients improves the accuracy and robustness of the supervised approaches used for face alignment applications. This work may be extended towards the development of accurate and fast eye gaze tracking systems.

Non-invasive Jaundice Detection using Machine Vision

Abstract The study investigated a non-invasive and instant method of jaundice detection using machine vision technique. Color images of sclera region of the eyes of healthy subjects and patients suffering from jaundice were acquired. Image processing algorithms were developed by using CIELab color model. The principal component analysis (PCA)-based discrimination analysis was applied over the color data obtained from patient’s sclera region, which showed a variance of 89%. The results of PCA biplot indicated correlations among jaundice patients and color attributes. Based upon these results, neuro-fuzzy-based software was developed for the prediction of jaundice as well as the calculation of degree of its severity. The experimental results show satisfactory performance as compared to the conventional chemical methods. The proposed technique is totally non-invasive and low cost.

Unobtrusive head gesture based directional control system for patient mobility cart

This work is based upon the design and development of novel technique which involves real-time identification of the head gestures using camera. The system incorporates a unique directional control system for maneuvering patient cart based upon the commands generated by head gestures. The work involved design and development of customized patient cart, camera arm attached along with camera, and tablet computer. The gesture detection was done by face alignment technique developed using regression based supervised learning method. The algorithm efficiently detects roll, and pitch under unconstraint conditions such as natural head movements and variable illuminations which avoid any loss of performance and sensitivity. The system is unobtrusive in nature as the patient cart is controlled by using normal head gestures only with a remotely fixed camera at a distance from the subject. Ultrasonic sensors based collision warning detection was also incorporated for safety purposes. The system is tested in laboratory environment and its performance was found to be excellent and most helpful for the effective empowerment of quadriplegic patients who want to maneuver their mobility cart themselves by head gestures only.

Intelligent remote vital health parameter monitoring system

This work involved development of a system which wirelessly transmits the vital health parameters of a patient to the remote physician accurately. The sensor network wearable to the patient measures the vital health parameters such as temperature, pulse rate, saturation of oxygen in blood, etc. The data acquisition system works constantly in both the hospital or home environments without any need of any care givers presence. The data is transmitted wirelessly over the network in a well synchronized manner to the physicians desktop in user friendly manner with a facility for recording up to the desired limit of time fixed by the physician. In emergency situations where any of the vital parameters of the patient was found be abnormal the warning signal was also generated in the form of SMS message to the doctors or physicians mobile phones immediately. This helps in decline of loss of lives by persistent examination by the doctor to a greater number of patients.

An Accurate and Simple Approach to Detect Eye Centers in Low Resolution Face Images

ABSTRACT This paper introduces a new approach for accurate eye center localization based upon optimized image gradients algorithm. The proposed approach requires pre-fetched feature descriptors to detect the accurate iris centers amongst the possible eye center candidates computed by image gradients over the face image data-set. The results of the proposed algorithm for detecting eye, iris, and iris centers over the test set images of publicly available low resolution and challenging data-set show an accuracy percentage of 98.9, 95.7, and 89.2, respectively. The comparison results obtained by the proposed approach were at par with the state-of-the-art techniques involving complex calculations and training requirements. So, the superior performance and outcome of the proposed approach show the usefulness of optimizing the results of simple image gradients by pre-fetched Scale Invariant Feature Transform feature descriptors in detecting iris centers under unconstrained environments. The proposed approach may be useful for the development of real-time eye gaze tracking application with improved robustness and accuracy.

Real-time stereo generation for surgical vision during minimal invasive robotic surgery

This paper proposes a framework for 3D surgical vision for minimal invasive robotic surgery. It presents an approach for generating the three dimensional view of the in-vivo live surgical procedures from two images captured by very small sized, full resolution camera sensor rig. A pre-processing scheme is employed to enhance the image quality and equalizing the color profile of two images. Polarized Projection using interlacing two images give a smooth and strain free three dimensional view. The algorithm runs in real time with good speed at full HD resolution.

Comparative analysis of unsupervised eye center localization approaches

The paper focuses on the comparative analysis of the latest and the most effective techniques for automatic eye center localization developed in recent years. The basic aim of the work is to identify the most accurate and efficient technique for eye center detection based upon an unsupervised approach. These techniques are developed using basic principles laid down by the various researchers without any optimizations, pre-processing or post processing steps and are explored over the periocular images extracted from standard database. The study showed varying processing times of these basic techniques for automatic detection of eye pupil over similar image dataset. The aim of this work is to show the significant and efficient unsupervised technique for eye center localization. The results of this study may be useful for the development of efficient and accurate approaches for eye center localization in eye gaze tracking applications.