BF-ASIFT-2DPCA and ABF-ASIFT-2DPCA for Face Recognition

详细信息    查看全文

• 作者：A. Vinay^a ; ; Joella Regina Mathias^b ; Anita M. Fathepur^b ; K.N. Balasubrmanaya Murthy^a ; S. Natarajan^b
• 关键词：ASIFT ; BF- ASIFT-2DPCA ; ABF-ASIFT-2DPCA ; ASIFT-2DPCA
• 刊名：Procedia Technology
• 出版年：2016
• 出版时间：2016
• 年：2016
• 卷：25
• 期：Complete
• 页码：411-419
• 全文大小：591 K

文摘

Walk into a shopping mall, the clerk will know your name and other details. The mall's cameras will known your identity by indexing you up in both their gallery and a wide gamut of marketing database they you may have subscribed to. Customers profile information including name, address, sale interests are displayed to the clerk. Further the clerk can infer what sort of sales pitches you’re most vulnerable to and how profitable a customer you are to that shopping mall. Similarly, walk by a policeman, they will identify your name, address and criminal record, if any. The underlying technology here is automated face recognition. This paper proposes two variations of the Affine-SIFT technique for face recognition. It outlines the applications of two filters namely the Bilateral filter and the Adaptive Bilateral filter. The quality of the image is enhanced before performing feature extraction and matching, so that matching will be more accurate. The use of filters not only increases the image quality, but also leads to more accurate key point matches which are crucial for image matching. The filters aid in pre-processing the image before feature extraction by smoothening the image, while preserving the edges and resulting in sharper images. This paper provides a comparative study of the various filters that can be used for pre-processing, some resulting in complete blurring of the image, some preserve the edges and some sharpen the image resulting in improved matches. It also explains how these pre-processing techniques impact the key point descriptors. The methodology also implements the Two Dimensional Principal Component Analysis (2DPCA) technique for dimensionality reduction of key points. The proposed model shows invariance to different levels of scale, illumination and poses to some extent.