Adaptive thresholding via Gaussian pyramid
C.K. Lee, C.H. Li · 2002
Automatic thresholding is an essential step in many applications of image analysis and pattern recognition. For highly noisy images whose statistics are not known a priori, the maximum entropy method is used to estimate the threshold for segmentation. Higher order estimation of the threshold which also takes into account the spatial distribution of the gray levels can also be used. However, these methods are difficult to implement and the computational requirements, in terms of speed and memory space, often exceed the hardware capability. Hence, the authors illustrate the application of Gaussian pyramid to reduce the image into a manageable size while preserving most of the content of the image. They also demonstrate the use of high level software, MATLAB, to aid the implementation of these complex algorithms.>