Barrel-type distortion compensated Fourier feature extraction

Abstract

Fourier based feature extraction is a common and powerful technique used in texture classification. In case of endoscopic imaging, often significant barrel-type distortions affect the feature extraction. Although images can be rectified using distortion correction techniques, in previous work feature extraction proved to suffer not just from geometric distortions, but also from effects within distortion correction. Distortion correction in combination with Fourier features has not been investigated so far. We introduce and evaluate three strategies to partially or completely compensate the geometric distortion in combination with Fourier features. With two methods, the interpolation within distortion correction can be omitted which should lead to a benefit in classification. Instead of making a general statement, we distinguish between certain frequencies and identify the positive and the negative aspects of the strategies. © 2013 Springer-Verlag.