基于图像处理提高木材识别准确性的新方法

Improvement of Wood Identification Accuracy Based on Image Processing

为了提高机器识别木材的准确性,从木材图像预处理的角度出发,以复原木材图像纹理细节为目标,提出了基于SCN-MSE的木材图像超分辨率重建方法。将高分辨率图像经离散小波变换,把4个子带中相应位置、大小相同的碎片组成模块,再利用局部二值模式的邻域强度建立训练数据库;将低分辨率图像中的低频子带碎片,与数据库同类模块中的低频碎片进行比较,通过去除领域中心的均方差,寻找相似度最大的高频碎片,保留低分辨率图像的低频子带;再经小波逆变换得到超分辨率图像。选取樟子松及其树皮为识别对象,将基于SCN-MSE的超分辨率重建图像与经过传统预处理图像,利用SVM多项式核函数进行识别。识别结果表明,本研究提出的方法提高了樟子松及其树皮的识别率。

In order to improve the accuracy of the machine recognition on wood, a SCN-MSE based method of wood image super-resolution reconstruction was proposed aiming to restore the texture details of the wood image. The high resolution images were transformed by discrete waves. Blocks were set up by the fragments which were in the corresponding positions and with the same sizes in four sub-bands. A training database based on nearest intensity-local binary pattern was then built. Discrete wavelet transform was used to extract sub-bands from low-resolution images. Regarding the low-frequency sub-bands as low frequency part, subtraction of center from neighbors-MSE was applied to obtain the high-resolution image block in the database,which was corresponding to the low-resolution image block,to reconstruct high frequency part. Then, inverse discrete wavelet transform was employed to generate super resolution images. Mongolica wood and bark was selected as object for recognition, SCN-MSE based super-resolution reconstructed images and the images pre-treated by traditional methods were subjected to SVM identification. The results showed that the proposed method improved the identification rate of Mongolica wood and bark.