针对火灾图像纹理识别问题,提出了基于Gabor小波变换的ICA火灾图像纹理识别算法,并根据火灾图像纹理识别特点进行了优化;首先用不同尺度和方向的Gabor滤波器对待识别图像滤波,得到其特征图像,然后将特征图像转化成特征向量作为ICA的输入,得到基矢量子空间,再将测试图像经过Gabor滤波器的特征向量投影到ICA子空间中得到系数向量作为目标识别特征,最后用支持向量机进行识别;通过与Gabor滤波器法和ICA方法的对比实验,表明该算法可以在火灾纹理图像的识别率上比传统方法提高5%以上,为火灾图像识别提供了一种新思路.
For the problem of fire image recognition,an algorithm combined with ICA and Gabor wavelet transform has been proposed for fire texture recognition,and optimized recognition method according to the fire image texture.Firstly,the image to be recognized is filtered by Gabor filters with different scales and orientations,and the characterized images is obtained.Then eigenvectors of these images are treated as the input of ICA.Base vector subspace can be obtained by using high-order statistic characteristics of ICA.Then the eigenvectors of test image filtered by Gabor filter are projected into ICA subspace,the coefficient vectors are treated as target recognition characteristics.At last,recognition is done by Support Vector Machine.After compared with Gabor filter method and ICA method,it indicated that proposed method can enhance recognition rate of fire texture images by over 5%.It gives a new approach for video fire detection.at robustness of this scheme gets a large progress,especially for the attacks of geometry crops and mosaic.