摘要
针对图像纹理分割,提出了采用图像Gabor多通道特征进行融合聚类方法.首先采用Gabor小波对图像进行卷积滤波,得到每个像素点的多尺度多方向的Gabor特征,然后对其进行标准化以及Gauss平滑,减少噪声影响.对每个优化后的Gabor特征作为训练值,采用融合聚类算法每次随机选择部分特征进行聚类,通过运行多次基聚类,然后对聚类结果采用投票的方式得到最终的图像纹理分割,通过人工合成纹理与自然纹理图像实验证明该方法对纹理的分类具有较高的正确率.
For the purpose of image texture classification, this paper proposes a clustering ensemble method by the use of multi-channel Gabor features. Firstly, it filtered the image by the multi-frequently and multi-direction Gabor wavelets, and then it standardized the image and Gaussian smoothed it to reduce the effects of noise;sec-ondly, it used the multi-channel filters as training data, randomly selected some of the feature for k-means clus-tering as the base and run the procedure for many times;lastly, it run clustering ensemble algorithms that voting for the base clustering results to get the final image texture classification. Experimenting with synthetic texture and natural texture images, it proves that the method of texture the classification has a high accuracy rate.
出处
《韶关学院学报》
2015年第2期6-10,共5页
Journal of Shaoguan University
基金
国家自然科学基金(F030402)
