文档下载 文档下载导航
设为首页 | 加入收藏
搜索 请输入内容:  
 导航当前位置: 文档下载 > 所有分类 > IT/计算机 > 计算机软件及应用 > P-Saliency Detection--A Spectral Residual Approach
侵权投诉

P-Saliency Detection--A Spectral Residual Approach

pothesisthatremovesredundanciesinthesensoryinput.Abasicprincipleinvisualsystemistosuppressthere-sponsetofrequentlyoccurringfeatures,whileatthesametimekeepssensitivetofeaturesthatdeviatefromthenorm[9].Therefore,onlytheunexpectedsignalscanbedeliveredtolaterstagesofprocessing.

Fromtheperspectiveofinformationtheory,effectivecodingdecomposetheimageinformationH(Image)intotwoparts:

H(Image)=H(Innovation)+H(PriorKnowledge),H(Innovation)denotesthenoveltypart,andH(PriorKnowledge)istheredundantinformationthatshouldbesuppressedbyacodingsystem.Inthe eldofimagestatistics,suchredundanciescorrespondtostatisticalinvariantpropertiesofourenvironment.Thesepropertieshavebeencomprehensivelydiscussedinliteraturepertain-ingtonaturalimagestatistics[4,25,17,18].Nowitiswidelyacceptedthatnaturalimagesarenotrandom,theyobeyhighlypredictabledistributions.

Inthefollowingsections,wewilldemonstrateamethodtoapproximatethe“innovation”partofanimagebyremov-ingthestatisticalredundantcomponents.Thispart,webe-lieveisinherentlyresponsibletothepoppingupofproto-objectsinthepre-attentivestage.

2.1.Logspectrumrepresentation

Oftheinvariantfactorsofnaturalimagestatistics,scaleinvarianceisthemostfamousandmostwidelystudiedproperty[20,17].Thispropertyisalsoknownas1/flaw.ItstatesthattheamplitudeA(f)oftheaveragedFourierspectrumoftheensembleofnaturalimagesobeysadistri-bution:

E{A(f)}∝1/f.(1)Onalog-logscale,theamplitudespectrumoftheensem-bleofnaturalimages,afteraveragingoverorientations,lies

approximatelyonastraightline.

Althoughthelog-logspectrumistheoreticallymaturedandhasbeenwidelyused,itisnotfavoredintheanaly-sisofindividualimagesbecause:(1)thescale-invariancepropertyisnotlikelytobefoundinindividualimages;(2)thesamplingpointsarenotwell-proportioned,thelowfre-quencypartsspansparselyonthelog-logplane,whereasthehighfrequencypartsdrawtogether,sufferingfromnoise[25].

Insteadofthelog-logrepresentation,inthispaper,weadoptthelogspectrumrepresentationL(f)ofLogspectrumcanbeobtainedbyL(f)=logandlogspectrum animage.A(f)representa-

.Thecomparisonbetweenlog-logtionisshowninFig.1.

Thelogspectrumrepresentationhasbeenusedinaseriesofliteraturepertainingtostatisticalscene

P-Saliency Detection--A Spectral Residual Approach

analysis

Source Image

y

P-Saliency Detection--A Spectral Residual Approach

tiy

tissnneettnnII ggooLL10

Frequency

Log frequency

10

10

Source Image

P-Saliency Detection--A Spectral Residual Approach

P-Saliency Detection--A Spectral Residual Approach

P-Saliency Detection--A Spectral Residual Approach

y

y

ttisnisneetntnII gogoLLFrequency

10

Log frequency

P-Saliency Detection--A Spectral Residual Approach

10

10

Figure1.Examplesoflogspectrumandlog-logspectrum.The rstimageistheaverageof2277naturalimages.

P-Saliency Detection--A Spectral Residual Approach

Figure2.Examplesoforientationaveragedcurvesoflogspectra.Thesecurvessharesimilarshape.Thelogspectrumiscomputedfromdown-sampledimage.Thesizeofeachlogspectrumis64×64

[22,23,21,11].Inthefollowingsection,wewillexploitthepoweroflogspectruminsaliencydetectiontasks.Ex-

上一页第2页

你可能喜欢

热门文档

相关文档

站点地图 | 文档上传 | 侵权投诉 | 手机版
本站所有资源均来自互联网,本站只负责收集和整理,均不承担任何法律责任,如有侵权等其它行为请联系我们.
文档下载 Copyright 2013 doc.guandang.net All Rights Reserved.  -  email