Operators for extraction of local information are essential components in an image processing system. This paper concentrates on the design and evaluation of convolution kernel sets enabling easy estimation of local orientation and frequency.
Consideration of interpolation properties and the limiting effects of the uncertainty principle leads to the definition of an "i deal" quadrature filter function. An optimization procedure is utilized to produce pairs of convolution kernels which implement an approximation of the desired function. A number of optimization results are presented.
To evaluate the performance of the optimized kernels in an image processing task, a series of experiments have been carried out. Examples are given of local orientation and frequency estimates for images with different signal to noise ratios. An angle deviation measure is defined and avector averaging scheme is introduced to increase angle estimation accuracy. Using a OdB SNR testimage, orientation estimates are produced having an expected deviation of less than 7 degrees.
1982. , 61 p.