基于Contourlet林火图像多重分形分割的研究

Forest fire image segmentation based on multifractal and Contourlet transform

提出了一种多重分形和Contourlet变换相结合的森林火灾图像分割新算法。先对图像进行Contourlet多层分解,得到一系列多尺度、局部化、多方向的子带图像。然后对低频子带进行多重分形特性分析,再将多重分形谱作为特征参量进行像素筛选,筛选出标志不同景物的像素点。最后,根据这些像素点的不同分布进行图像分割,从而实现烟雾的分割。实验结果表明,这种算法能够有效地实现火焰和烟雾的分割,提高了森林火灾图像分割的有效性。

A new segmentation algorithm for forest fire image based on muhifractal and Contourlet transform is proposed. The algorithm firstly decomposes the image into several subband images with multi-scale, localization and multi-direction using Contourlet transform. Then, muhifractal analysis of low frequency sub-band is performed. The muhifractal spectrum is used as the characteristic parameters to filter the pixels that mark different scenery. At last, according to the distribution of these pixels, fire and smoke segmentation is realized. Experimental results show that the algorithm can segment flame and smoke effectively and improve the efficiency of segmentation of forest fire image.