Research On An Algorithm Based On Middle Semantic Representation Of Image Scene Classification

As the growing numbers of images and vedios，how to use computer to analysethe image imformations automatically becomes an important problem.However thereare so many classes of image and the dimensionof the features that extracted fromimages are always very high，it is very hard to do images classification automaticallyby computer. It is clearly that the massive picture information than manual labor canbe done. For this reason, many researchers have been studied on how to do imageclassification automatically by computer vision.In images classification task，the image scene classification is an importantprimary problem. The key problem of scene classification is how to use appropriatesemantics to describe the images.We present a local constrains sparse coding spacepyramid matching(SPM) model to solve problems which in semantics describing inour praper..First we use a coding scheme called locality-constrained sparse coding todecripe the low-level features，then get the final representation of a image by maxpooling with SPM.At last we use SVM to do classification.As the aspect ofcodebook，we also improved the traditional method.And use a effective method to geta adaptive codebook.This overcomplete codebook can descripe the image featuresbetter.The experimental results show that our locality-constrained sparse coding withSPM can achieve better results on many scene image database.