A Curve Fitting Based Image Segmentation Method

In this paper a method is proposed to segment the board in Snooker games automatically. The algorithm involves the following steps: first, separable second derivative of Guassian steerable filters, which are linear combination of basis filters and can implement filtering with variant direction, are used to extract the edges in both horizontal and vertical directions. The input image f(x,y) is convolved with every base filter, then weighted the summation of outputs of all the base filter, the resulted image g(x,y) can be obtained. Second, the candidate board regions are extracted by color segmentation. After image dilation, which can make the points which are around the background merge to the objects, it is helpful that extracting the edge pixels exactly. And next, the candidate edge pixels are extracted by both the peak value of edge pixel and the candidate board region. Finally, the true edge pixels can be obtained by curve fitting, and the board is segmented by the results of curve fitting.