基于多光谱图像融合和形态重构的图像分割方法

一些成熟的瓜果果实在单一的光谱图像中,果与叶的灰度值只存在微小差异,常用的图像分割方法不足以把果与叶区分开,为此,提出一种基于多光谱图像融合的形态学重构分割方法.首先,采集同一目标的可见光彩色图像和近红外图像,对此多光谱图像分别采用主成分分析(PCA)、小波变换以及可见光图像H分量与近红外图像NIR的算术组合(NIR/H)等方式进行融合处理:然后,对融合图像进行形态学重构分水岭分割.多幅苹果和番茄图像的同标提取试验结果表明,对可见光图像和近红外图像的PCA和小波变换融合图像进行形态学重构分水岭分割,可以得到较好的分割效果,尤其是小波变换融合图像的形态学重构分水岭分割效果更具有自适应性.

Image segmentation method based on multi-spectral image fusion and morphology reconstruction

The color values of some mature fruits are the approximations with those of their branches and leaves. The gray values of fruit and leaf have only the small difference. The fruit target cannot be extracted precisely, from the leaf background using the single spectral image. A novel image segmentation method was proposed based on multi-spectral image fusion and morphology reconstruction. First, some fusion methods were applied to the multi-spectral image, such as the principal component analysis (PCA) fusion, the wavelet transformation fusion as well as arithmetic combination (NIR/H) fusion of the visual and the near-infrared images of the same target; then, segmentation algorithm based on the morphology reconstruction was applied to the fusion image. Through the experiment from many apple and tomato images, the results indicate that the morphology reconstruction segmentation based on multi-spectral image fusion, gives better segmentation results. In particular, the effect of morphology reconstruction segmentation using watershed transformation based on wavelet transformation fusion, is the best with good self-adaptability.