Design team member: Lesley Phord-Toy
Supervisor: Dr. David A. Clausi
Background
The human image processing and recognition system is a very sophisticated one, and has yet to be even approximately duplicated by artificial means. Many image identification tasks require human operators to extract useful data and cannot be done with computers alone. An area in image analysis that is of particular interest involves the analysis and classification of textures.
A texture gives the impression of depth and surface orientation, as well as giving cohesion to individual objects, while providing boundaries between objects within an image. The human eye can discriminate between different textures within a 2-dimensional image and people can be trained to identify what they represent in the 3D world. This process can be called texture analysis and classification.
Although image processing and pattern recognition is a large field of study, there is still much to be learnt in many of its areas. Despite the substantial amount of work that has been done in providing artificial image processes for texture, there are no clear solutions in the area of texture analysis and classification.
Project description
This workshop will explore several image processing and pattern recognition techniques and their relation to texture analysis and classification. A methodology that will specify by which these techniques can be compared and contrasted will be designed based on the findings of the research and experimentation for this project. If encouraging findings result from this work, the methodology will be able to suggest appropriate texture feature extraction methods as well as accompanying classification techniques for varying types of images.
It is hoped that this workshop will provide some useful information in the field of texture analysis and classification, which can then be used in further work in this area. This work could then potentially lead to real-world applications such as automatic or semiautomatic analysis of remotely sensed data including sea ice imagery.
Design methodology
The first half of this workshop was concentrated on gaining familiarity and knowledge of texture feature extraction techniques by implementing the following techniques:
Co-occurrence
Probability
Matrices.
Markov
Random
Fields.
Dyadic
Gabor
Filter
Banks.
The methods implemented in this workshop will be used to formulate a strategy and methodology for their use. The code implementing these methods will not be used for processing a huge amount of images for classification. Therefore, it was decided that the efficiency of the rate of implementation of these techniques was of greater importance than the efficiency of the actual code needed to run them. As a result, MATLAB was used to implement the methods to take advantage of its substantial mathematical functionality as well as its image processing toolkit.