煤矿井下早期火灾图像中火焰区域、火焰余辉及非火焰高灰度干扰区域三者的灰度值十分接近,利用传统的Chan-Vese (CV)模型很难将火焰区域精确地提取出来.针对这一问题,提出了一种改进的CV模型以实现煤矿井下早期火灾图像的精确分割.在计算目标和背景区域拟合中心时,引入自适应权值进行加权平均,充分考虑了像素点灰度值与拟合中心的差异,并据此确定该点对拟合中心的贡献度,更加精确地计算目标和背景区域的拟合中心;为了加速模型的演化,引入曲线内外区域像素的中值绝对差,替换模型中的内外区域能量系数,提高模型分割效率.最终达到快速提取早期火灾图像中火焰区域的目的.大量实验结果表明,与现有的Otsu算法、CV模型、引入能量权重的CV模型、引入梯度信息的CV模型以及两种类似提出模型的CV模型相比,利用改进CV模型对煤矿井下早期火灾图像,能取得更好的分割效果,并且满足实时性要求.%Because of a great similarity in grayscale value among fire region,fire after glow,and non-fire region interference with high grayscale value in the early mine fire image,it is hard to extract fire region by traditional CV model accurately.To overcome this problem,an improved CV model was proposed to achieve the accurate segmentation of early mine fire image.When calculating the target and background fitting centers,the adaptive weights were introduced to weight fitting centers.It fully considered the grayscale value differences between pixels in each region and fitting centers and the contribution of pixels to calculating the fitting centers was determined according to the grayscale value differences.Therefore,the object and background region fitting centers can be obtained more accurately.In order to accelerate the evolution of the proposed model,the median absolute differences of the pixel grayscale values inside and outside the curve are incorporated,which can adaptively adjust the region energy weights inside and outside the curve,instead of original region energy weights,to improve the segmentation efficiency.Finally,the fire region of the carly mine fire image is extracted accurately and efficiently.The proposed method was compared to the Otsu algorithm,he CV model,the CV model incorporating energy weight,the CV model incorporating the gradient information and two CV models incorporating similar weights like the proposed method.The extensive experiment results show that the improved CV model can gain much a better segmentation accuracy than other methods,and satisfy real-time requirement.
展开▼
