Big Data-Driven Feature Extraction and Clustering Based on Statistical Methods

摘要

Big data-driven feature extraction is a challenging process because it contains a variety and voluminous of data. But, in the current scenario of the Internet and the multimedia data-driven necessitates handling of complex data. Nowadays, it becomes a significant challenge to the Internet-based service provider to store voluminous data. To overcome this difficulty, this article provides a novel technique for big data-driven feature extraction, based on statistical methods. At first, the proposed method preprocesses the given input key-frame, that is, normalizes and removes noise. The noise-removed key-frames are separated into background scenes and forefront objects; features are extracted from the background scenes and forefront objects. The extracted features formulated as a feature vector. To validate the extracted features that whether it correctly represents the specific frame or similar frames, the feature vector is associated with the feature vectors in the feature vector catalogue. The proposed feature extraction method matches and retrieves the frames from the video database. It yields average correct retrieval rate of 95.29 per cent. The results obtained from experiments show that the proposed feature extraction method gives the average retrieval precision of 95.29 per cent. The enactment of the proposed feature extraction method is analogous to the existing methods.