emd算法与图像频分析-模式识别与智能系统专业毕业论文.docx

垒!!!!竺! ．．一一—— EMD AIgor．thm and Image Time·Frequency AnaIysis Cui Feng (Pattem Recognition afld Imelligent Systems) Institute ofAutomation，Ch血ese Academy of Sciences Directed by Peng Silong Abstract The time—frequency analysis for image signalis one of the signmcant and difflcult domains in image pmcessing techn0109y．The aim of it is to analysis how the frequency components Vary in local time or local space and utilize the time—fkquency features， such as instantaneous行equenc弘to accomplish image processing assignrnents．The t00ls aIld algorithms for image time—frequency analysis are just researched in t}1is thesis． This thesis researches EMD and quatemionic spectmm anaIysis deepl)‘which are novel time·疔equency a11alysis tools，and achieves actual ev01ve in image analysis At first，tlle 2一D separable EMD based on the characters of 1一D EMD is introduced in this mesis，which extends me 1’D EMD to 2-D；Then，in orderto aJlalysis analylic signal for 2一D s培nals，this t11esis researches the quatemionic analytic signal and imroduces a fast quaternionic Fourier transforIIl and inverse quatemionic Fou“er transfo咖 algorittlIll． Simultaneously， this thesis puts fonv盯d a novel phase definition for quatemionic aIlal”ic signal，which oVercomes s。me shortcomings of nle exjst defjnjtion． In the application domain of image time一行equency analysis，this lhesis also does some research． First， an llIlsupenrised texture segmentation algorithm is proposed in this mesis，based on the characters of 2-D separable EMD and quatemionic analytic signal； Second， this thesis introduces a novel texture compression metllod based on 2·D separable EMD；At last，it analysis textufe W01d —111．  EMD AIgorithm and Image Time-Frequency Analysis decomposi“on in quatemionic spectnml domain， based on fast quatemionic spectmm compming algoritllm． Theory a11alysis a11d eXperiment results indicate that the algoritllms pmposed in this也esis have respectiVe advantages，such as irHlovation，accuracy，simplic咄 a即licability， etc．Thi