Robust multi-scale orientation estimation: Directional filter bank based approach

摘要: Orientation estimation is considered as an important task in many subsequent pattern recognition and image enhancement systems. In a noisy environment, the gradient-based estimator provides poor results. A pre-smoothing Gaussian function with appropriate scale conventionally used to get improved gradients. Later on, family of functions range scales employed for estimation, this referred multi-scale orientation estimator. To provide groundwork comparison, more formal framework based on scale-space axioms, spatial domain presented. Then improvement purposes Fourier approach, where directional filter bank (DFB) structure embedded framework, proposed. This DFB approach. The paper presents comparison work local orientations using approaches both domain. Fourier-domain two linear combinations are deployed, one across image, other scales. opposed only combination scales, simple techniques. Further more, DFB-based approach extracts best by comparing contrasting all possible their respective strength measures. measure method variance, free from inaccurate gradient calculation. Simulations conducted over test images well real fingerprints. Our objective results indicate that always yields better estimates at variable level noise compared stand alone approaches. improvements made estimate can largely be attributed use double bands