Fern 알고리즘에서 효율적인 특징 정보 구성 방법

Abstract

In target recognition field, Fern algorithm has been researched because of the high-recognition accuracy and the simple structure. Feature vector of Fern are constructed randomly. Consequently target recognition accuracy is depended on randomness. This paper proposes the efficient method to construct feature vector for Fern. Firstly, the proposed method calculates a correlation coefficient between feature vectors and then uses a correlation coefficient for measurement about the uniformity of distribution of feature vector. we use 2bit binary pattern to feature vector. We present through experiment result the relation between the uniformity of distribution of feature vector and target recognition accuracy.