Image Compression Using Sub-Block Singular Value Decomposition Method for Automatic Object Recognition

カテゴリ: 部門大会

論文No: MC1-6

グループ名: 【C】平成19年電気学会電子・情報・システム部門大会講演論文集

発行日: 2007/09/04

タイトル(英語): Image Compression Using Sub-Block Singular Value Decomposition Method for Automatic Object Recognition

著者名: MD. FOISAL HOSSAIN (University of the Ryukyus),Mohammad Reza Alsharif (University of the Ryukyus),Katsumi Yamashita(Osaka Prefecture University)

著者名(英語): Md. Foisal Hossain (University of the Ryukyus),Mohammad Reza Alsharif (University of the Ryukyus),Katsumi Yamashita(Osaka Prefecture University)

キーワード: Image Compression|Singular Value Decomposition|Sub-Block|Rank Reduction|Lossy CompressionAutomatic Object Recognition

要約(日本語): The focus of this paper is to present a lossy approach to compress digital images where the recovered images have sufficient quality to be used for computer vision tasks such as skin detection, pattern recognition etc. For image compression, we use Singular Value Decomposition (SVD). In this research we investigate the effect of applying SVD with different block size for image compression and also for recognition. The use of SVD allows us to produce three matrices of the digital images, two of them are unitary matrices and the other is a diagonal matrix with the same dimension of the original image. We reduce the rank of the diagonal matrix. The use of the resulting singular values allows us to represent the image with a smaller set of values and achieve lossy image compression. The compressed method preserves useful features of the original image to perform Automatic Object Recognition (AOR) over the recovered image. For AOR we use software of National Instrument called “Vision Builder”. The use of smaller block size with SVD results in higher compression with smaller space although recognition rate does not reduce too much. We show the preliminary results of the system with different block size for compression and for AOR.

PDFファイルサイズ: 5,268 Kバイト