【摘要】 小波变换具有良好的“时间—频率”局部化特性及多尺度变焦距特性对于二维图象的小波变换，其梯度模值提供了图像的边缘信息在大尺度时，图像边缘稳定，但存在有位移小尺度时，边缘定位精确，但易受噪声影响噪声和边缘都具有较高的空间频率噪声的能量小，在大尺度下，其小波变换系数值小边缘的能量大，在大尺度变换下，其小波变换系数值大由多尺度小波变换系数的变化情况，估计边缘的类型采用多尺度小波变换系数作为四分树结构的分开—合并法图像分割的一致性度量从而在大的图像块中，去除噪声的影响，在小的图像块中，以小波变换的局部极大值精确定位边缘，根据边缘信息进行分开—合并法图像分割实验表明，算法得到满意的结果更多还原

【Abstract】 The wavelet transform has good "time-frequency", localizatisn feature and multiscale variable lens character. The abstract value of gradation gives the information of image edge. When in high scale, the image edge is stable, with displacement ;while in low scale, the image edge is located very well,with subtle to noise.Both the edge and noise have high frequencies, the noise has low energy, so the coefficient of wavelet transform in high scale is small. The edge has large energy, the coefficient of wavelet transform in high scale is large. We can estimate the kinds of edge by changing the wavelet transform coefficients in serial scales. Be using the wavelet transform coefficients as the measure of uniform of four-tree structure of image segment and in big image block, we can eliminate the noise and in small image block we can locate the edge very well. Then we can use the information of edge to segment image by the split-and-merge algorithm. The experiments show that this approach can get good result.更多还原