비음수 행렬과 텐서 분해를 이용한 지역적 특징 추출과 유사도.

In this paper, non-negative matrix factorization (NMF) and non-negative tensor factorization (NTF) are used to express the local features of an object as basis vector (or basis matrix). To recognize an object through its partial appearance, it should be expressed as a combination of basis vectors (or basis matrices). We applied the part-based image representation to the vehicle recognition for autonomous driving, where occlusion is a problem. As a result, the recognition rate was about 90% which shows that the part-based representation is effective. We also propose an local similarity based on the relative ratio between vectors and showed that it is more robust against occlusion than conventional similarities such as Euclidean distance and cosine similarity. At the 60% of occlusion, the conventional similarity showed a recognition rate of 33.3%, while the local similarity showed a recognition rate of 66.7%, which means that the local similarity is superior and less sensitive to the loss of feature information due to occlusion. [ABSTRACT FROM AUTHOR]