The paper presents a novel approach to extract a semantic image object based on an optimized Harmonious Adaboost algorithm

shortly HAB

which produces less generalization error and high performance compared to the Gentle Adaboost Algorithm. Some key techniques in the proposed schema

including the pre-processing of image character

the training of object detector and the extracting of semantic image object

are discussed. The experiment shows that the recurrent training process improves the performance of the object detector

and the extracting results demonstrate the availability of the work.