关键词: 图像分割, 直方图降维, 阈值选取, 最大类间方差法

Abstract: The issue of poor resistance to noise and unreasonable is pointed out based on
two-dimensional histogram regional straight points method. A new threshold segmentation method is
proposed, and the calculation formula of the method is deduced. Firstly, in this method, noise
interference weakened for threshold′s segmentation through the reconstruction of two-dimensional
histogram based on detailed analysis of noise distribution in the two-dimensional histogram, and then,
the region division is transfered from eight partitions into two partitions in two-dimensional
histogram. Thus the two-dimension search space of threshold is reduced to one-dimension. Finally,
simulation results of existing two-dimensional histogram segmentation algorithm and our method are
given respectively. Theoretical analysis and experimental results show that our method could be used
in nearly all the two-dimensional histogram threshold segmentation, especially in threshold
segmentation with the contaminated image. It makes the inner part uniform, the edge accurate in the
threshold image and has better tolerance capability to noise. The running time is significantly
reduced.

Key words: image segmentation, histogram dimensionality reduction, threshold selection, Otsu
algorithm