Multiscale analysis for image enhancement with steerable filter

Image processing involves multi-dimensional manipulations. One of the main image contents is edge-like feature, such as line and boundary, of which the orientation is an important property. Thus the processing module should be able to preserve this kind of information and suppress noise. This paper concentrated on the fingerprint image enhancement, where a novel approach is proposed by using steerable filter. We utilize the steerability and orientation selectivity of the filter bank to estimate the local orientation of ridges' and use the orientation variance to calculate the scales, upon which the enhancement result can be obtained adaptively from the filtering results. Then a simple segmentation method is used for binarization. Experimental results show that the proposed approach works well for images in different scales and qualities.