Method for accurately extracting image foregrounds based on neighbourhood and non-neighbourhood smooth priors

Disclosed is a method for accurately extracting image foregrounds based on neighbourhood and non-neighbourhood smooth priors, which mainly comprises the following steps: interactively marking foreground points and background points; for each unmarked pixel of an input image, initializing the alpha value thereof by using a colour sampling method and calculating the confidence level thereof, and for a pixel of which the confidence level is greater than a given threshold, adopting the alpha value thereof as a known pixel; for each pixel of the input image, calculating a data item weight, a neighbourhood smooth constraint item weight and a non-neighbourhood smooth constraint item weight of each pixel, so as to construct graphical models of all the pixels of the input image; and according to the alpha value of the known pixel, under the constraint of the graphical models, solving the probability of each pixel point belonging to the foreground through the energy minimization equation, so as to obtain a mask image (alpha matte). The present invention has good performance in the aspects of accuracy and mask image continuity calculations and so on, and can be widely applied and popularized to the field of video program making and so on.