基于增量特征和局部奇异性的水下图像分割法

Underwater image segmentation using local regularity and increment feature

针对深海三类典型的纹理:热液、岩石和海水,提出了一种新的基于增量特征和局部奇异性的水下图像分割方法.定义了一种新的增量特征s(d),反映像素点由小尺度到大尺度变化的剧烈程度;利用二维小波变换分析图像局部奇异性,并结合多项式拟合的方法提取奇异性特征P;最后根据s(d)和P组成特征矢量,结合k均值聚类方法,对图像进行分割.实验结果表明该方法能有效地分割出上述三类纹理.

There are basically three kinds of veins （stones, sea water and hot liquids） in deeply oceans. A new increment feature parameter s（d） is proposed which could describe the variations of point-wise. Local regularity feature P was obtained, using 2D wavelet transformation and polynomial curve fitting. A new feature vector composed of s（d） and P was given, together with K-means clustering, a segmentation approach was given based on local regularity and increment feature which could fully extract the texture information and find the right classification results for three kinds of the underwater veins. Experimental results indicate that the algorithm could reduce the clustering errors and improve the convergence, having better segmentation quality.