Adaboost modular tensor locality preservative projection: face detection in video using Adaboost modular‐based tensor locality preservative projections. Issue 7 (3rd May 2016)

• Record Type: Journal Article
• Authors: Rajeshwari, Janthakal; Karibasappa, Kwadiki; Gopalkrishna, Madigondanahalli Thimmaiah
• Is Part Of: IET computer vision. Volume 10:Issue 7(2016)
• Page Start: 670
• Page End: 678
• Publication Date: 2016-05-03
• Subjects: tensors -- learning (artificial intelligence) -- face recognition -- image classification -- computer vision -- feature extraction -- video signal processing -- image sequences; Adaboost modular-based tensor locality preservative projection -- automatic face detection -- computer vision -- pattern recognition applications -- subspace method -- visual learning -- illumination variation -- pose variation -- occlusion variation -- block-based tensor locality preservative projection -- AMTLPP algorithm -- block-based TLPP algorithm -- feature extraction -- weak classifier -- strong classifier -- YouTube celebrity -- McGill face dataset -- video sequence collection; Computer vision -- Periodicals; Pattern recognition systems -- Periodicals; 006.37
• DOI: 10.1049/iet-cvi.2015.0406
• Languages: English
• ISSNs: 1751-9632

Abstract

Abstract : Automatic face detection is a challenging task for computer vision and pattern recognition applications such as video surveillance and traffic monitoring. During the last few years, subspace methods have been proposed for visual learning and recognition which are sensitive to variations in illumination, pose and occlusion. To overcome these problems, the authors have proposed a method that combines block‐based tensor locality preservative projection (TLPP) with Adaboost algorithm which improves the accuracy of face detection. In the proposed algorithm Adaboost modular TLPPs (AMTLPPs), the face image is divided into overlapping small blocks and these block features are given to TLPP to extract the features where TLPP take data directly in the form of tensors as input. AMTLPP algorithm selects the optimal block features from the large set of the block features which forms the weak classifiers and are combined to form the strong classifier. A number of assessments are conducted for YouTube celebrity, McGill face dataset and also on collected video sequences of an own dataset recorded under indoor, outdoor, day, sunset and crowded environment. Experimental results show that the proposed approach is effective and efficient.