The automated understanding of textual information in images is an important problem to solve for the Computer Vision and Document Analysis for extracting that information for processing. This needs to generate required word regions and the remaining to be filter out the nontext area. For this, we extract the connected components (CCs) in images by using the maximally stable extremal region algorithm. Whereas in the existing system the region based method is considered. These extracted CCs are partitioned into clusters so that we can generate candidate regions Instead of using heuristic rules for clustering we train an AdaBoost classifier which determines the adjacency relationship and cluster those CCs by using their pair wise relations. Then we normalize candidate word regions and determine whether each region contains text or not. Adaboost classifier is based on multilayer perceptrons and we can control recall and precision rates with a single free parameter we develop text/nontext classifier for normalized images. Finally we obtain the extracted text by matching the trained set of templates.