THE SELECTION OF NATURAL SCALES IN 2D IMAGES USING ADAPTIVE GABOR FILTERING

This paper analyzes how the natural scales of the shapes in 2D images can be extracted. Spatial information is analyzed by multiple units se nsitive to both spatial and spatial-frequency variables. Scale estimat es of the relevant shapes are constructed only from strongly respondin g detectors. The meaningful structures in the response of a detector ( computed through 2D Gabor filtering) are, at their natural level of re solution, relatively sharp and have well-defined boundaries. A natural scale is so defined as a level sigma producing local minimum of a fun ction that returns the relative sharpness of the detector response fil tered over a range of scales. In a second stage, to improve a first cr ude estimate of the local scale, the criterion is also rewritten to di rectly select scales at locations of significant features of each acti vated detector.