Channel based Threshold Segmentation of Multi-Class Cervical Cancer using Mean and Standard Deviation on Pap Smear Images

Cervical cancer is the second most dangerous metastatic tumor where grows in a woman's cervix. Pap smear is the simplest screening test for the detection of cancer at the starting stage. Cervical cancer has two types of Normality and abnormality cancer which contains the cell and cytoplasm in the same structure. It is difficult to identify a cancerous nucleus in the cell. In Image processing, various algorithms are used to segment the nucleus alone in microscopic images. The primary scope of this paper is focusing on Pre-processing, Segmentation and Feature Extraction with six levels of Pap images. The performance evaluation has been calculated based on the segmentation results. In the first phase, pre-processing used mean filters to remove noise and enhanced with CLAHE. In the second phase, Segmentation used threshold value by taking the sum of three channels with the proposed methodology in mean and standard deviation. In the third phase, Properties of image regions, Shapes, Textures and some statistical features are extracted after segmentation. To evaluate performance measure used SSIM for each type of cancerous segmentation that is compared with K-means and Fuzzy C-means algorithm. Thus, the proposed work of separation and addition of RGB channel based segmentation gives the best results for nucleus segmentation in Pap smear images. The accuracy level 94.60 % has been obtained by using SVM and KNN classification for 182 Pap smear images with a class of six labels. Matlab R2016a is used as a programming tool.