Mengjie Zhang, Fast Object Detection

Fast Object Detection using Genetic Programming and Neural Networks

  • What is object detection and why is it so important?

    As more and more images are captured in electronic form the need for programs which can find objects of interest in a database of images is increasing. For example, it may be necessary to find all tumors in a database of x-ray images, all cyclones in a database of satellite images or a particular face in a database of photographs.

    This image shows an example of a problem of this kind. It shows a human retina. We are required to find all of the micro aneurisms and haemorrhages, as indicated by the white squares. (Note: The picture is presented at too coarse a level of resolution for the difference between micro aneurisms and haemorrhages to be evident). In this image we are required to find all occurrences of the head side of the 5 cent coins.

    More examples are: sonar target recognition, missile classification, tank and helicopter recognition, face detection, handwritten digit and character recognition, bacterial growth detection and also many medical applications.

  • Tasks and Goals of this project:

    The goal of this project is to investigate an efficient, learning/adaptive approach to multiclass object detection problems based on pixel statistics. Pixel statistics are domain independent, pixel level image features, which can be easily extracted even selected from a preexisting library. Instead of using traditional vision methods, this project will focus on learning and adaptive approaches such as genetic programming and neural networks. Specifically, the project will investigate:

    • Typical good examples in large images for the learning process;
    • Good fitness functions in GP for object detection; and
    • Fast training algorithms.
    A sequence of image data sets providing object detection problems of increasing difficulty will be used as the test bed, such as face detection problems and object detection in human retina images.

  • More information can be seen from Meng's research projects . Related Publications can be viewed from here: Meng's recent publications. Contact me if you want to get a copy of those papers, or if you want to know more detail about this project.