In order to address the issue of gray scale inhomogeneity and initial contour sensitivity

an image segmentation model based on multi-scale local feature is proposed. Different from traditional local neighborhood defined in square shape region

the circular shape is used to capture more local information in the model. Taking into account intensity varies in different levels in local region

the method combines multi-scale structure with mean value filter is proposed to acquire multi-scale local grayscale information. An approximation of true image

which is obtained by transforming grayscale inhomogeneity model

is incorporated into the local Gaussian distribution fitting（ LGDF） model and the energy function is constructed with multi-scale local feature. The theoretical analysis and experimental results demonstrate that the proposed method can rapidly and accurately segment grayscale inhomogeneity image

and also has strong robustness to the initial contour since multi-scale structure weakens the influence of intensity inhomogeneity.