A new vision approach for local spectrum features in cervical images via 2d method of geometric restriction in frequency domain

• Authors: Viara Van Raad
• Venue: CVBIA'05 Proceedings of the First international conference on Computer Vision for Biomedical Image Applications
• Year: 2005

Digital colposcopy is an emerging new technology, which can be used as adjunct to the conventional Pap test for staging of cervical cancer and it can improve the diagnostic accuracy of the test. Computer aided diagnosis (CAD) in digital colposcopy has as a goal to segment and outline abnormal areas on the cervix, one of which is an important anatomical landmark on the ectocervix – the transformation zone (TZ). In this paper we proposed a new method for estimation of the local spectrum features of cervical cancer in vivo. We used a 2D method to estimate the energy of the local frequency bands, using a geometric restriction (GR). In the current work we reported up to 12 dB difference between the local power spectral density content of the region of interest (ROI) and (ROI)complimentary for the mid-frequency band. We devised a method to present pseudo-color visual maps of the cervical images, useful for CAD and successful ROI segmentation.