Generation of Compact and Effective Training set for Image Database Classification

Abstract

In today s digital world, huge amount of images and videos are easily generated, accessed and shared. Before exploring and analyzing these images, it would be better to organize them in meaningful categories. So there is a need to make an automatic classifier which will classify these images according to their visual contents. newlineClassification is an important preprocessing step for content-based image retrieval (CBIR) system, especially when thousands of images are involved. There are two major steps in supervised classification system. Initially feature vectors for all training images are generated. This will be considered as the training set. Second step is to build the classifier using the training set. Accuracy of the classification system depends on many factors. The quality and the size of training set are the important factors. This work mainly focuses on the Generation of training set for classification. The original contribution to knowledge is the generation of an efficient and compact set of training feature vectors from given set of training images. Training and testing set of images are two disjoint sets. Features are extracted from images in transform domain. Since attention is on the first step of supervised classification system, the classifier is build using simple nearest neighbor (NN) classification. newline