Face recognition from surveillance video by rank-level fusion using interval type-3 fuzzy sets and deep learning features

Face recognition from surveillance video is challenging and critical as it offers potential solution to the unlawful activities within the secured areas. Often, surveillance video captures sequence of activities of subjects without their prior knowledge. In this article, we propose a face recognition system from surveillance video using fuzzy rank-level fusion utilizing interval type-3 fuzzy sets and deep learning features. Unlike normal ranks, it generates fuzzy ranks using different convolutional neural networks (CNNs) for a face image to be recognized. In particular, it utilizes deep learning features, outputs (confidence factors) of the CNN, and within-class and between-class specific information to generate interval type-3 based fuzzy ranks. This mechanism mitigates the classification uncertainty arises due to within-class dissimilarity and between-class similarity. Finally, these fuzzy ranks, generated by different CNNs, along with the respective confidence factors are convolved to yield the final fuzzy ranks. The experimental results on the Chokepoint dataset confirm its efficiency as compared to existing rank fusion techniques.