图像直方图局部极值算法及其在边界检测中的应用

利用多项式拟合方法获得的直方图轮廓的连续函数估计,解决了拟合函数的次数的选择问题。通过设计拟合精度标准,可以根据直方图的性质自动确定拟合函数的次数。该函数不仅克服了伪极值点对实际极值点的干扰,而且扩大了常见的各种极值算法的应用范围。实验结果表明,拟合效果与多项式的次数无正比例关系,而且当多项式次数加大时,拟合函数两端会出现震荡现象。这种方法对图像的噪音具有很强的鲁棒性,并可以利用连续函数的性质方便地获得函数的局部极值,从而为图像自动二值化方法提供了条件。

Algorithm for local extremum of histogram and its application in the boundary detection

Histogram is the most important statistic property of images, especially in those algorithms and analysis based on the threshold for images. Due to the discrete property of histogram, it is difficulty to attain the position of local extremum by common methods. The polynomial estimation is proposed to get outline of histogram. The mainly problem to be solved is the proper selection of index of polynomial. The index can be auto-selected by the fitting criterion. This function reduces the interferes of pseudo-extremum and enlarges the applications of extremum algorithm existed. The experiment shows that the accuracy standard is not proportional with the index of polynomial function. The two ends of the graph would have a concussion phenomenon, if the index grew. This method has stronger robust to the noises in the images. We can get the local extremum with the continuous property of this fitting function. This is also the basis of automatic two-value methods.