Implementation of an object recognition algorithm using normalized mutual information

A study has been done on implementing an object recognition algorithm utilizing similarity metric. The algorithm measured highest similarity values between images in a database as available on the internet. The main objective was to implement an algorithm using the metric of normalized mutual information and fully developed in Java, supported by an image processing architecture, ImageJ. The main part of this work was the image comparison process, based on a non-segmented method where the similarity in images was measured utilizing all the image intensity values specified by a region of interest on the images. Assumptions were made for the implementation of the algorithm after considering possible object recognition problems and constraints encountered in the real situation to retrieve the stolen item images. The algorithm was tested for 100 images of a self-constructed image database and produced the output of sorted ranks of highest similarity measure in the form of tabbed-separated data and a text file.