基于权重系数的木材图像增强及识别

Wood Image Enhancement and Recognition Based on Weight Coefficient

为了进一步提高机器识别木材的准确性,提出基于权重系数的木材图像增强方法。将采集的图像利用离散小波变换分为4个不同的子带,对LL子带采用权重系数检测相似的像素点,对LH、HL和HH子带利用局部方差法进行定向筛选,对上述子带均采用傅里叶系数作差法,同时结合阈值进行混淆判断;对LL子带采用方向自适应滤波消除混淆,对LH、HL和HH子带采用自适应小波收缩消除混淆,最后经过小波逆变换得到增强后的重建图像。选取榆木木片和树皮为识别对象,对其图像进行增强处理,然后利用BP神经网络进行识别。结果表明,本研究方法较传统的图像增强方法提高了榆木木片和树皮的识别率。

In order to further improve the accuracy of machine recognition of wood species, a wood image enhancement method was proposed based on weight coefficients. The collected images using discrete wavelet transform were divided into four different subbands,in which LL subband was used to detect similar pixels with weight coefficient, the left three, i. e. LH, HL and HH subbands were applicable to the directed screening by partial variance method. All the four subbands were subjected to Fourier coefficient difference method, and the confusion were judged combined with threshold values. Adaptive directional filter was use to eliminate the confusion in LL subband,while the confusions in LH,HL and HH subbands were avoided by using adaptive wavelet shrinkage method, and to obtain reconstructed enhanced images through inverse wavelet transform. Elm wood and hark were selected as the identify objects. Images were treated by using the methods proposed, and then recognized by BP neural network. The experimental results showed that the proposed methods were better than the traditional image enhancement method in the recognition rate of elm wood and bark.