Abstract
In recent years, object recognition which appears in the face recognition in a digital camera has been studied extensively. The learning-based object recognition methods find common characteristics from a large number of learning sample data. These methods work well when the images in the learning data set have a unified appearance, such as face. However, the methods are not effective against objects which change in appearance. In this study, we propose new learning-based object recognition technique to the appearances of a object in the training data set. The learning data are classified according to their appearances. Thus, the classifier with each high precision is built by learning. And, a robust classifier could be built in the appearance of the object by integrating each built classifier into 1. In the experiments, we show the results of a performance comparison using the proposed method and a conventional method.
