ROTATING-KERNEL MIN-MAX ALGORITHMS FOR STRAIGHT-LINE FEATURE ENHANCEMENT

The rotating-kernel min-max transformation is a nonlinear image-proces sing operation that can be applied to the enhancement of directional f eatures in noisy images. Associated with a particular transformation a re (a) a convolution kernel and (b) a function that maps to a final ou tput value the maximum and minimum values measured at point (x, y) in the convolution output as the kernel rotates through 360 degrees. Freq uently used kernels are narrow in one direction and broad in the other , typically with rectangular, triangular, or Gaussian profiles in the long direction. Simple but effective functional mappings include I-out (x, y) = [Max(x, y) - Min(x, y)] and I-out(x, y) = [1 - [Min(x, y)/Max (x ,y)](m)]. Improved results are often obtained if successive rotatin g-kernel min-max transformation operations are performed in cascaded s ystems. Two binarization procedures based on the rotating-kernel min-m ax transformation can be used to extract straight-line features from n oisy gray-scale images. The effects on the processed image of kernel t ype and size, mapping function, and binarization scheme are discussed.