Hand detection has many important applications in human-computer interaction. But hand detection is a difficult problem because hand image can vary greatly in images. Vision based hand interfaces require fast and extremely robust hand detection. Large data sets are needed in the process of creating classifiers to detect. This study proposes an alternative method for creating positive images that the classifier needs. This method, which is to be presented, is aimed at obtaining a large number of positive images autonomously from a certain number of hand images, instead of annotating positive images under human supervision. Therefore, less time have been spent and a wider set of data has been achieved.