Online video object classification using fast similarity network fusion

In this paper, we propose one online video object classification algorithm using fast Similarity Network Fusion (SNF). By constructing sample-similarity network for each data type and then efficiently fusing these networks into one single similarity network that represents the full spectrum of underlying data, SNF can efficiently identify subtypes among existing samples by clustering and predict labels for new samples based on the constructed network, which make it distinct in data integration or classification. The main problem of data online classification using SNF is its complexity. The proposed fast SNF (FSNF) in this work consists of two main steps: dividing the matrix into two parts and replacing the main part of testing matrix using the same part of training matrix. Since the main computation in SNF is to get the main part of matrix, this replacement can reduce most of the computation load. From the experiments based on online surveillance video object classification, it can be observed that: compared with SNF, the proposed FSNF can gain 16 times speed increasing with only 0.5%-0.6% accuracy losing; FSNF also significantly outperforms the existing traditional algorithms in classification accuracy.