Fields of Research
Our principal research area has been the development of techniques to extract and represent objects contained in images. A secondary research area has been the application of Artificial Intelligence techniques to a number of search and recognition problems. The majority of our projects combine the outputs from both these research areas to produce commercial solutions.
Realistic Talking Head Models
In APIIT's first major project, Realistic Talking Head Models, an image compression technique based on Principal Component Analysis (PCA) was developed which allowed individual frames in talking mouth videos to be compressed down to as little as eighty bytes. Not only did the compressed image parameters capture all the perceptually important information in the original images, such as the lips, teeth and tongue, skin texture and shading; but, they also showed smooth variation with time. Images compressed using this technique were used to construct hidden Markov Model (HMM) based audio-visual synthesis
engine, which was subsequently commercialised as Speech-Tutor®, a teaching aid for the deaf.
Signature Recognition Engine (SIROX)
APIIT's second major project was the development of SIROX, a commercial handwritten signature recognition engine for the banking industry. In this project a number of image processing techniques were developed to extract signatures from bank cheques and then parameterise them so that they could be compared against the customers' signatures in a bank's database. One requirement of the system was that it had to work when only one master signature was available. This meant that a number of standard techniques, as implemented in other competing commercial signature recognition products, were not applicable.
To solve the problem we used a number of non-standard techniques, such as Radon transforms, morphological skeleton analysis as well as developing some of our own.
Anti-Porn Solutions
APIIT's third major project was the development of a number of software applications for detecting and blocking pornographic images. In this project images are captured from the PC screen and analysed using a number of techniques such as skin-tone and texture analysis, face detection and brightness flow analysis to produce a set of features. The features are then processed using Neural Networks and Classification and Regression Trees to determine whether the image is pornographic or not.
iSMART
In APIIT's fourth major project, iSMART, a suite of resource planning and scheduling solutions for maximising resource utilisation in educational institutions was developed. At the heart of these solutions lie Genetic-Algorithm and fuzzy-logic based engines for planning schedules and assigning the necessary resources, such as classrooms and lecturers. Further, unlike other commercial timetabling products, iSMART engine has customisable constraint handling options which can be changed as the institute's underlying business logic changes.
Car License Plate Extraction and Recognition Technology (C.A.R.P.E.T)
APIIT's fifth major project, Car License Plate Extraction and Recognition Technology (C.A.R.P.E.T), is a solution for car parks, toll booths, traffic enforcement, traffic monitoring, stolen car tracking and access control. C.A.R.P.E.T utilises an array of the cutting-edge techniques such as i) variants of Otsu thresholding for image enhancement in poor lighting conditions, ii) Burn's algorithm for noise reduction, iii) morphology for license plate location, iv) variants of the drop-fall method for character segmentation, and v) neural networks for character recognition. The main advantage of
C.A.R.P.E.T is that, unlike most competing systems, it does not require special hardware such as proprietary video cards and vehicle sensors.
Back to top
