New Methods for Image De-noising and Edge Enhancement in Cervical Smear Images Segmentation

The segmentation of cytoplast and nucleus from a cervical cell image is one of important techniques for automatically detecting abnormal cervical cells. Noise on an image often makes segmentation inaccurate. This paper presents two new techniques, named trimming-mean filter and bi-grouping enhancer, to effectively eliminate noise and make the object boundaries more discernible. In this paper, the two proposed techniques are integrated with other techniques to create a nucleus and cytoplasm contour detector (NCCD detector) to automatically server the cytoplasm and nucleus from a cervical smear image. Compared to the gradient vector flow active contour model (GVF-ACM) and a texture-based segmentation method, the NCCD detector has a better performance in segmentation. Five commonly used performance criteria, including misclassification error (ME), edge mismatch (EM), region nonuniformity (RU), relative foreground area error (RFAE), and shape distortion penalty (SDP), will be taken to evaluate the segmentation techniques. The experimental results indicate that the NCCD detector is more effective in segmenting nucleus and cytoplasm from a cervical smear image.