建模连续视觉特征的图像语义标注方法.PDF

22 8 Vol. 22 No. 8 2010 8 Journal of ComputerAided Design Computer Graphics Aug. 2010 1, 2, 3) 1) 1, 3) 1) 李志欣 施智平 刘 曦 史忠植 , , , 1) ( 1001 0) 2) ( 541004) 3) ( 10004 ) ( lizx@ ics. ict. ac. cn) : , . (PLSA) , , ; , , PLSA , PLSA , 2 , . 5000 Corel , , . : ; ; ; ; : T P3 1 Semantic ImageAnnotation by Modeling Continuous Visual Features 1, 2, 3) 1) 1, 3) 1) Li Zhixin , Shi Zhiping , Liu Xi , and Shi Zhongzhi 1) ( K ey L abo rat or y of I nt el l ig ent I nf or mati o n P rocess i ng , I ns t i tut e of Co mp u ti ng T echno l og y , Chine s e A c ade my of S ci ence s, Bei i ng 1001 0) 2) ( Col l eg e of Comp ut er S ci ence and I nf or mati on T echno l og y , Guang x i N or mal Univ er s it y , Gui li n 541004) 3) ( Grad uat e Univ er s it y of Chi nes e A c ade my of S ci ence s, Bei i ng 10004 ) Abstract: In order to bridge the semantic gap in image retrieval, this paper proposes an approach to annotate image automatically by modeling continuous visual features directly. Firstly , w e extend probabilistic latent semantic analysis ( PLSA ) to model continuous quantity. In addition, corresponding ExpectationM aximization algorithm is derived to determine the model parameters. Secondly, in terms of the characteristics of different modalities, we present a semantic annotation model w hich employs continuous PLSA and standard PLSA to model visual features and textual w ords respectively . The model learns the correlation betw een these tw o modalities by an asymmetric learning approach and then it can predict