频域内基于权重系数的木材图像增强及识别

发布时间：2018-11-13 09:54

【摘要】：在木材行业中,最常用的识别木材的方法以视觉观测为主,但是利用计算机进行识别更为准确。在图像采集过程中由于环境以及采集设备的限制因素,导致人们所采集到的图像并不理想,因此会使得提取的木材中的纹理信息丢失,这样就会导致在利用图像进行木材图像的识别时出现误差。本文现有的图像增强算法的基础上,提出了基于权重系数的木材图像增强算法,分析了低高频、高低频和高高频等波段的小波变换的性能,从而有针对性的消除图像中的混淆部分,提高图像质量。论文研究的主要内容包括:1.本文重点介绍了直方图均衡化算法、直方图规定化算法、低通滤波算法、高通滤波算法和小波变换算法这五种传统的图像增强算法理论。2.在现有的图像增强算法的基础上,提出了基于权重系数的木材图像增强算法,此方法主要是对图像中混淆备份的定向筛选,即将整幅图像分为LL、LH、HL和HH四个不同的波段,对LL子带进行归一化处理。并利用计算权重系数的方法获得相似模块,同时结合自适应滤波模块进行混淆检测;对LH、HL和HH子带采用局部方差法进行混淆检测。最后将其利用方向自适应小波收缩进行混淆消除,通过小波逆变换完成混淆消除恢复图像。3.将基于权重系数的木材图像增强算法与直方图均衡化算法、直方图规定化算法、低通滤波算法、高通滤波算法等方法进行实验对比,并对所得实验结果进行主观和客观分析。4.将基于权重系数的图像增强算法应用到木材图像识别中,以榆木和榆木树皮作为识别试样,分别用基于权重系数的木材图像增强算法和传统的图像增强对榆木木片和榆木树皮图像进行预处理,并用BP神经网络对图像进行识别,且对其识别结果进行分析和比较。
[Abstract]:In the timber industry, the most commonly used method of wood identification is visual observation, but it is more accurate to identify wood by computer. In the process of image acquisition, because of the limited factors of environment and acquisition equipment, the image collected by people is not ideal, so the texture information of extracted wood will be lost. This will lead to errors in the recognition of wood images using images. Based on the existing image enhancement algorithms, a wood image enhancement algorithm based on weight coefficient is proposed, and the performance of wavelet transform in low high frequency, high and low frequency and high frequency bands is analyzed. Thus the obfuscation part of the image is eliminated and the image quality is improved. The main contents of this paper are as follows: 1. This paper focuses on five traditional image enhancement algorithms: histogram equalization algorithm, histogram specification algorithm, low-pass filtering algorithm, high-pass filtering algorithm and wavelet transform algorithm. 2. On the basis of existing image enhancement algorithms, a wood image enhancement algorithm based on weight coefficient is proposed. This method is mainly used for directional filtering of obfuscation backups in images, that is to say, the whole image is divided into four different bands: LL,LH,HL and HH. The LL subbands are normalized. The similarity module is obtained by calculating the weight coefficient, and the aliasing detection is carried out by combining the adaptive filtering module, and the local variance method is used to detect the LH,HL and HH subbands. Finally, it uses directional adaptive wavelet shrinkage to eliminate confusion, and accomplishes the obfuscation elimination and restoration image by inverse wavelet transform. The wood image enhancement algorithm based on weight coefficient is compared with histogram equalization algorithm, histogram specification algorithm, low-pass filter algorithm, high-pass filtering algorithm and so on. The experimental results are analyzed subjectively and objectively. 4. The image enhancement algorithm based on weight coefficient is applied to wood image recognition. Elm and elm bark are used as the recognition samples. The wood image enhancement algorithm based on weight coefficient and the traditional image enhancement algorithm are used to preprocess the elm and elm bark images, and the BP neural network is used to identify the images, and the recognition results are analyzed and compared.
【学位授予单位】：内蒙古农业大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP391.41

【参考文献】