Abstract: This paper presents a self\|adaptive algorithm based on prediction, introduces the theory of the algorithm and gives many examples and comparisons. The algorithm computes the mean and std. diff. of current and previous pixels' density to estimate the next point's density range: if the next point's density falls in the range, it is not a boundary point; otherwise, it may be a boundary point. The results show that the algorithm of image boundary detection is more effective than the others.