On Scale and Orientation Adaptive Filtering
Leif Haglund, Hans E. Knutsson, Gösta H. Granlund · 1992
This paper contains a presentation of a scale and orientation adaptive filtering strategy for images. The size, shape and orientation of the filter are signal controlled and thus locally adapted to each neighbourhood according to an estimated model. On each scale the filter is constructed as a linear weighting of fixed oriented bandpass filters having the same shape but different orientations. The resulting filter is interpolated from all scale levels, and spans over more than 6 octaves. It is possible to reconstruct an enhanced original image from the filtered images. The performance of the reconstruction algorithm displays two desirable but normally contradictory features, namely edge enhancement and an improvement of the signal-to-noise ratio. The adaptive filtering method has been tested on both real data and synthesized test data. The results are very good on a wide variety of images from moderate signal-to-noise ratios to low, even lower than 0 dB, SNR.