基于频域增强和差分盒维数的早期火灾烟雾图像分割

根据烟雾的半透明性特征和分形性质,提出一种基于频域增强和差分盒维数的烟雾图像分割算法。首先,利用小波变换空-频域的特性,通过对图像多层分解后小波系数的加权处理得到烟雾纹理增强图像;然后,运用差分盒维数方法遍历图像,计算出各像素分形维数值,由阈值法得到烟雾分割图像。最后,通过形态学膨胀运算使分割图像更加完整。实验结果表明,该算法能有效利用小波频域增强的特点,减小烟雾薄弱区内背景的影响,使该区域烟雾的分形维数更多地集中于阈值内,提高了烟雾分割的准确性。

Early Fire Smoke Image Segmentation Based on Enhancement of Frequency Field and Differential Box-Counting Fractal Dimension

According to the translucent characteristics and fractal property of smoke, an algorithm of smoke seg-mentation was proposed based on enhancement of frequency field and differential box-counting（DBC） fractal dimen-sion. Firstly, smoke texture enhancement image was obtained by dealing with the weighted wavelet coefficients of image multilevel decomposed with the characteristics of wavelet transforms’space-frequency field. Then, every pix-el fractal dimension was calculated by using the differential box-counting method to traverse the image and smoke segmentation image was obtained by threshold. Finally, the image segmentation was more completed by using dila-tion algorithm. The experimental results show that the algorithm can utilize the features of wavelet transform effec-tively in the frequency field enhancement , reducing the impact of background in smoke weak region, making smoke fractal dimension of the region more focused on the range of the threshold, improving the accuracy of smoke segmen-tation.