Research Of Cervical Cell Image Classification Based On Feature-united PCANet

Cervical cancer is increasingly threatening the health of women,then early s creening and prevention of cervical cancer is very necessary.The artificial error of cervical cell image interpretation can be efficiently decreased by automatic c omputer-aided diagnosis,as well as labor cost.Then cervical cancer screening tech nology can be extended rapidly,which has good social value and economic valu e.In this paper,some key technologies of cervical cell image classification and recogniti on are studied,which include cervical cell image denoising,cervical cell image en hancement,cervical cell image feature extraction and cervical cell classification and recognition.The main research content are as follows:(1)An image denoising method,which uses non-local self-similarity prior learning is used for cervical cell image denoising based on patch group.The res ult of simulation experiment shows that the denoising method,which is used in t his paper,can remove noise of cervical cell image effectively and protect structu re information of cervical cell image.Furthermore,the good robustness of this de noising method is showed,because of PSNR and SSIM decreased slowly when n oise of cervical cell image increase.(2)An image enhancement method,which is called Bi-histogram Equalizati on with adaptive sigmoid function,is used for cervical cell image enhancement,e specially emerging image feature,then image feature can be extracted easy.(3)In this paper,the united-feature PCANet,which is based on PCANet,is proposed for feature extraction of cervical cell image.The difference between uni ted-feature PCANet and PCANet is the network structure.The united-feature PC ANet unites output feature of interface layer with output feature of the last laye r as the last output feature,which can reduce losing image feature in some degr ee and the differences between images can be represented.The SVM is used for cervical cell image classification when images features are extracted.The results of simulation experiment are as follows,the two classification accuracy of cervi cal cell image is 95.71%and three classification accuracy is 85.40%,which sho ws some application value.(4)The system of cervical cell image classification and recognition is des igned based on MATLAB GUI,which includes training module and detecting mo dule.The function of this system is training united-feature PCANet and detectin g cervical cell image.Functions of the system operate well and simply.In general,this system has a good application value.