Design of Gabor wavelets for analysis of texture features in cervical images

The trends in the digital technology opened possibilities for digital colposcopy systems to be equipped with software for computer-aided diagnosis as an adjunct method to the traditional Pap-smear test for cervical cancer prevention. In this paper we demonstrate a method,which outlines the area of metaplastic changes, known as the transformation zone (TZ), via Gabor filter analysis. We propose a new frequency domain design scheme for Gabor wavelet with varying tuning frequencies and orientations, specifically designed to reduce the redundancy in the wavelet-based presentation of the cervical image. The scheme calculates the Gabor wavelets either at dyadic or at continuous scales. We demonstrate, that we can achieve close to 80% accuracy in discrimination on the region of interest (ROI), using that particular Gabor wavelet scheme. The results show that the method can be used effectively as an adjunct computer-aid in order to increase the accuracy of the test, targeting the primary input of the traditional cervical test - the sampling procedure.