Preprocessing Techniques for Image Analysis Applications
Hong Zhang, Atlantic State University, Georgia, hong@drake.armstrong.edu
The preprocessing of data (e.g., normalization, feature extraction, and custom
kernel function) before they are presented to a learning machine is
often crucial in developing an effective machine learning system. Even
though modern machine learning methods such as the support vector machine
have the built-in mechanism of regularization, preprocessing remains
an important tool to facilitate effective training, improve inductive bias,
and incorporate prior knowledge. It is especially useful for problems with
large input dimensions and limited sample sizes such as the applications
involving image analysis. The raw image data are not a desirable form
of input for learning machines. Preprocessing is necessary in virtually
any non-trivial problems of image analysis. In this talk, examples
of the preprocessing techniques and algorithms relevant to various image
analysis applications are discussed. Topics include statistical methods,
clustering, image
processing techniques, and special invariant kernels.