:: Gopal's Home Page ::
Gopal Datt Joshi

PhD Scholar
CVIT, IIIT Hyderabad

Continuous effort--not strength or intelligence-- is the key to unlocking own potential.


Automatic Retinal Image Analysis for Glaucoma Detection

Glaucoma is a chronic optic neuropathy that often leads to blindness if inadequately treated. Glaucoma is emerging as the second most common cause of blindness in India and the world. A vast number of persons with glaucoma in India remain undetected due to inadequate access to eye care services and lack of awareness. Early detection and appropriate institution of therapy and periodical monitoring of individuals is crucial since the cost of treatment has shown to significantly escalate with increasing severity of the disease.
The goal of this project is to investigate the potential of automated image analysis for detecting glaucoma from retinal (fundus) images. The objective is to develop algortihms to detect and quantify changes in the patterns of the optic disc with respect to its normal appearance, which is vital in the diagnosis of glaucoma.

DrishtiCare: A tele-screening platform powered with retinal image analysis

The basic objective of DrishtiCare is to help in building an ecosystem for early diagnosis of eye disorders caused by diabetes. It would bridge the gap between the requirement and the availability of retinal examination services for diabetic patients. The main aim is to provide affordable, simple, and easily accessible retinal examination services to such patients. It involves a tele-screening delivery platform combined with a set of Computer Aided Diagnosis (CAD) system based tools to provide a solution in terms of accessibility and scalability. We intend to leverage the capabilities of Image processing techniques to significantly enhance the potential of our platform as compared to traditional tele-ophthalmology solutions. This work is supported by TPDUP: Technopreneur Promotion Programme (TePP), Ministry of Science & Technology, Department of Scientific & Industrial Research.

Diabetic Retinopathy Detection from 'Colour' Retinal Images

The aim of this project was to develop a set of tools to analyse colour fundus images (CFI) for detecting and tracking image structures relevant to diabetic retinopathy based on computer vision techniques. The algorithms and architecture developed here can be used to support screening programmes for the early detection and monitoring of diabetic retinopathy; evaluation of treatment outcome; telemedicine for distance education and treatment.

Diabetic Retinopathy Detection from 'Fluorescence' Retinal Images

This project was aimed at developing computer-aided tools for the detection and profiling the proliferation of structures of interest in Fluroscene fundus angiogram (FFA). The structures of interest that has been identified are microanuerysms (MA) and capillary non-perfusion (CNP), the early and late signs of Diabetic Retinopathy, respectively. Image processing techniques were developed to detect and quantify these structures.

paperDiff: Finding Differences between document images

This work was carried out during my stay at HP labs Bangalore, India. The aim of PaperDiff was to compare two printed (paper) documents using their images and determine the differences in terms of text inserted, deleted and substituted between them. This lets an end-user compare two documents which are already printed or even if one of which is printed (the other could be in electronic form such as MS-word *.doc file). The proposed method for realizing PaperDiff is based on word image comparison and is even suitable for symbol strings and for any script/language (including multiple scripts) in the documents, where even mature optical character recognition (OCR) technology has had very little success. PaperDiff enables end-users like lawyers, novelists, etc, in comparing new document versions with older versions of them.

Automatic detection of Age-related Macular Degeneration (AMD) from Colour Retinal Image

This project was aimed at extracting drusen from colour retinal images. Drusen are yellow coloured lipid material that are deposited underneath the retina. They are considered as the precursor of advanced age-related macular degeneration (AMD). Drusen may vary in shape, size, colour and other morphological properties. These variations make drusen detection a challenging task and conventional detection and segmentation techniques fail to work. We have developed a model based method for detecting and grading drusen using curvature information. Preliminary quantitative and qualitative analysis performed at lesion-level showed promise in the approach.

Script Identification from Document Images

This work was aimed to identify a script class in a given document image. This facilitates many important applications such as automatic archiving of multilingual documents, searching online archives of document images and for the selection of script specific OCR in a multilingual environment. Here, we modeled script identification as a texture classification problem and examined a global approach inspired by human visual perception. A generalised hierarchical framework was proposed for script identification. A set of energy and intensity space features for this task were also presented. The framework serves to establish the utility of a global approach to the classification of scripts. The framework was tested on two datasets: 10 Indian and 13 world scripts. The results demonstrated that the framework can be used to develop solutions for script identification from document images across a large set of script classes.

Boundary Detection from Natural Images

Boundary detection in natural images is a fundamental problem in many computer vision tasks. In this work, we have argued and showed that early stages in primary visual cortex provide ample information to address the boundary detection problem. In other words, global visual primitives such as object and region boundaries can be extracted using local features captured by the receptive fields. The anatomy of visual cortex and psychological evidences were studied to identify some of the important underlying computational principles for the boundary detection task. A scheme for boundary detection based on these principles was developed. Results of testing the scheme on a benchmark set of natural images, with associated human marked boundaries, showed the performance to be quantitatively competitive with existing computer vision approaches.

Checker Playing Robotic Arm

The objective of this project was to build an autonomous checkers playing robot which could interact with a human without the need of any third operator or any unnatural means of communication. It consists of three main sub-systems: vision sub-system; AI sub-system and Mechanical/robotics subsyetem. A web camera based video acquisition was used and moves on the checkers board was identified from the consecutive video frames by applying image change detection. The identified move was passed to the AI module in which checkers logic was written to compute robot's next move against identified move. For checkers logic, we have used standard AI techiques such as min-max and alpha beta pruning. Next, move given by the AI module was executed by the robotic arm which was taken from Lynxmotion company. Locomotions of robotic arm were computed using the geometry of the robot and degrees of freedom supported by it.
A human interface was later designed using Java Speech API which has basic level speech recognition and speech synthesis capabilities. A simple markup language and a very simple grammar-based recognition is designed for human-robot interaction.

Pioneer: A Path Finding Robot

The Pioneer, a path finding robot was our B-tech third year (2003-2004) project. The goal of our project was to develop a similar robotics system which can do a collision-free navigation in an environment. This robot was built using h/w components which were taken from a toy car. Detection of obstacles was performed using two IR sensors which were mounted on the front side of the robot. A set of navigation action rules were designed to handle different collision states. Entire logic and inter-component interaction procedures were implemented on a micro-controller. A one-sided radio frequency based connection with a remote terminal was established for the localisation of the robot. Despite of many practical difficulties, it was a satisfactory experience.