基于二进小波变换的多尺度图切割方法

摘要

Aiming at the default that when the traditional multi-scale ncut algorithm is applied to image edge detection, it only uses the Gaussian low-pass filtering and derivative derivation to extract contour and ignores the accuracy of contour, this paper presents that using dyadic wavelet transform to detect multi-scale edge. In the multi-scale weight matrix con-structing process, using the feature of each wavelet to solve similarity matrix, it greatly reduces the computational com-plexity to solve similarity matrix and reduces the memory consumption. Experimental results prove that proposed algo-rithm obtains better segmentation results than the traditional multiscale Ncut algorithm.%针对传统多尺度图切割方法进行图像边缘轮廓提取时仅采用高斯低通滤波和一阶微分求导，而忽略轮廓精度的缺陷，提出利用二进小波变换的方法进行多尺度边缘检测。在建立多尺度相似矩阵的过程中，根据每层小波的特性直接对各个尺度的相似矩阵进行构造，有效地减少了求解相似矩阵的运算复杂度，降低了内存消耗。实验结果表明，该方法与传统多尺度图切割方法相比具有更好的分割效果。