Abstract: The overall total of undermine video image is dark,the contrast of the information distributed in undermine video image is low,due to the large amount of dust existed in undermine,the undermine video image is polluted by the random noise with high density,so,the difficulties of image interpretation and analysis are increased.Firstly,the undermine video image is conducted multi-scale ridgelet decomposition,the low-frequency and high frequency coefficients are conducted inverse ridgelet reconstruction respectively,the background image and details image of the original undermine video image are obtained;secondly,according to the grey histogram distribution characteristics of the background image,the histogram specification method based on Gaussian distribution function is used to improve the visual effect of the background image;then,based on analyzing the characteristics of median filtering(MF),the boundary discrimination noise detection method(BDND) is adopted to improve the median filtering(MF),a improved median filtering algorithm based on BDND is proposed,and it is used to deal with the details image,the two adaptive threshold values of t₁ and t₂ are set by the algorithm proposed in this paper based on the grey values distribution features of the local region of the details image,the grey values distributed in the range of［t₁,t₂］ corresponding pixel points are regarded as pixel points that are not polluted by random noises,the other pixel points are processed by median filtering algorithm respectively;finally,the background image processed by histogram specification method and the details image processed by the improved median filtering algorithm are superimposed.The C++ language is used to write the computer programme of the improved median filtering algorithm,the experimental results show that the filtering effects of the undermine video image with different fuzzy degree of the algorithm proposed in this paper is superior to median filtering algorithm,weighted median filtering algorithm and switch median filtering algorithm.

Key words: Undermine video image, Ridgelet transform, Boundary discrimination noise detection method, Median filtering, Gaussian distribution function