关键词: 细胞学涂片, 计算机辅助诊断, 形状特征, 曲率

Abstract: Cell contour geometry is an important reference for smear interpretation, that has important significance for the study of computer-aided diagnosis of cervical lesions. In order to improve the robustness of the existing recognition method based on the geometry of the shape of the template matching, the paper proposes a new geometry feature extraction based on the geometric shape matching the curvature of the shape. The method proposed by comparing the template contour and contour curvature to be identified, computes the similarity between the curvature of the curve, resulting in characteristic cell shape contour. The paper by successively rotating the contour to select the best matching solves the problem of inconsistency contour direction, and the problem of inconsistent contour size by an area equivalent radius of a circle than a zoom magnification ratio contour method. The paper verifies that geometry feature extracted has a scale and rotational invariance, and the experimental comparison results with improved Hausdorff distance show that the proposed shape features can more accurately identify the cell contour geometry.

Key words: cytological smear, computer-aided diagnosis, shape feature, curvatures