Matthew Johnson

Research Interests

My research concentrates on the following areas:

  • Computer Vision
    • Object recognition
    • Image segmentation
    • Image retrieval
    • Local Image Descriptors
    • Construction of Image/Text Corpora
  • Natural Language Processing
    • Word Sense Disambiguation
  • Machine Learning
    • Decision Forests
    • Support Vector Machines
    • Boosting

Below is a list of my research projects:


Semantic Texton Forests

This project concentrates on the application of decision forests to perform efficient semantic segmentation of images, resulting in a pixel-level labeling of an image with object labels.

Semantic Texton Forests
Semantic PhotosynthesisSemantic Photo Synthesis

This project focuses on ways to leverage advances in computer vision to aid the user experience through intelligent interfaces, in this case for the creation of concept art and the searching of image databases.

Multi-Label Boosting

A system which is able to learn the correlations between image parts and multiple labels using a boosting framework for multi-label machine learning algorithms.

ML Boost

The ImCor dataset is a linking of semantically tagged and disambiguated data from SemCor to the Corel image dataset, used in my research with Kobus Barnard on word sense disambiguation with pictures.