Object Extraction Using Novel Region Merging and Multidimensional Features

Understanding natural images is a difficult task. One method to accomplish that can be, first, segment the image into regions of similar characteristics and then apply some object extraction scheme. Alternatively, extraction of characteristics of the desired objects can be initiated at the beginning. In this paper, we propose a scheme that adopts the former approach. An image is first segmented and then a novel region processing algorithm is applied. It is found that the proposed algorithm is capable of removing a high percentage of insignificant regions. Following region processing, a multidimensional feature extraction scheme is used. A set of five primitive and non-primitive features are used to create the feature vectors. The image dataset is divided into two parts, i.e. training and test sets. Results show the effectiveness of the proposed feature vector in extracting known objects present in natural and specific domain images.