Local Scale Control for Edge Detection and Blur Estimation

Selecting the appropriate spatial scale for local edge analysis is a challenge for natural images, where blur scale and contrast may vary over a broad range. While previous methods for scale adaptation have required the global solution of a non-convex optimization problem [8], it is shown that knowledge of sensor properties and operator norms can be exploited to define a unique, locally-computable minimum reliable scale for local estimation. The resulting method for local scale control allows edges spanning a broad range of blur scales and contrasts to be reliably localized by a single system with no input parameters other than the second moment of the sensor noise. Local scale control further permits the reliable estimation of local blur scale in complex images where the conditions demanded by Fourier methods for blur estimation break down.