Wall-pasted cell segmentation based on circular Gabor filter bank

Wall-pasted cell segmentation plays an important role in ectogenetic anti-virus experiment. Traditional Gabor filter (TGF) has been introduced to enhance the cell boundaries, which resulted in a reasonable segmentation. However, the boundaries are of arbitrary orientations, which make the orientation selectivity of TGFs become a computational burden, because Gabor filters of six or more orientations are needed to sufficiently sample the orientation domain. Moreover, although TGFs can enhance the boundaries, they are not suitable for enhancing or extracting the features of the wall-pasted cells. In this paper, circular Gabor filter (CGF) bank is newly proposed to extract the features of both the cells and the boundaries. CGFs of lower radial frequencies is suitable for extracting the features of the cells´ bodies, while that of higher radial frequencies can extract the features of the boundaries of different orientations, because CGF own the property of rotation invariant. Another priority of the CGF is that the filter design is easier than that of TGF, because there are fewer parameters to deal with. A popular Gabor filter bank design method is introduced and applied to the design of CGF bank, which is then used to extract the features of the wall-pasted cell. Satisfactory segmentation of the cells are obtained by clustering the feature vectors using K-means clustering, and the scheme is quite robust to the variances of the parameters of the CGF bank.