Analysis and Implementation of Feature Extractors for Content-Based Similarity Search (Bachelor Thesis, Finished)
This thesis gives a short introduction to content-based similarity search and a summary overview of image features that can be used for similarity searches. Two feature extractors, and the features they produce, are de- scribed in more detail: Color moments and texture Gabor moments. Both are global features based on statistical descriptions of the colors or textural patterns in the images. These two extractors have been implemented in Java for this work, along with three methods to divide images into regions for region-based features. A series of tests has been run to compare their speed performance, obtained feature values and retrieval results with those of an existing implementation in C++. The quality of the retrieval with these two implemented extractors has been measured against a gold stan- dard; a collection with manually assigned relevancy labels on the images, prepared for the ImageCLEF 2007 medical image retrieval challenge, was used for this.
Start / End Dates
2007/08/15 - 2007/12/15