关键词: 图像分割, 光滑样条函数, 林火识别问题, 阈值

Abstract: Abstract: Image Segmentation is an important and also a critical preprocessing step for image analysis, pattern recognition and computer vision. In this paper, for the special characters of the forest flame fire, based on the principle of smoothing spline, we propose an image automatic multi-threshold approach, termed as HistSplineReg (Spline Regression for Histogram). Essentially, HistSplineReg is a two-step method. Firstly HistSplineReg seeks a smoothing spline function to fit the 1D image histogram, and then automatically segments the given image by multiply thresholds from extremums of spline regression function. Compared to the existing multiply thresholding methods, the advantages of HistSplineReg lie in 4 folds: 1) it is simple and easy to study, which is quite consistent with the human intuition; 2) it is constructed on the smoothing spline, which has a solid mathematic theories; 3) both the number and the size of multiple thresholds can be automatic determined; 4) HistSplineReg can be analytically solved, and its computing consume mainly focus on the matrix Cholesky decomposition, while the size of matrix depends on the level of image pixel, rather than the scale of image; 5) it has only ONE trade-off parameter to balance the empirical error and regressor’s smoothness. Furthermore, for the forest fire recognition task, we also provide an experimental reference value. Finally, we show our experimental results on some forest fire digital images and report results in the following aspects: gray level image, all color channels, and the synthetic image from different color channels. Compared to SVR and Polynomial, in the viewpoint of data fitting, the HistSplineReg achieves more clear segmentation images. We experimentally conclude that the red channel merits more attention for the RGB-mode forest fire image among foresaid three methods.

Key words: Image segmentation, smooth spline function, forest flame fire recognition, threshold