基于图像分割和加权Fisher判据的彩色IC图像降维

Dimensionality reduction in color IC images based on image segmentation and weighted Fisher criterion

在处理彩色集成电路(IC)图像的过程中,为了降低计算的复杂度,经常需要将彩色图像转换成灰度图像后再进行处理。本文将数据降维中优化判据的思想引入彩色图像到灰度图像的转换中。为了求得最优降维方向,必须寻找一个判据来衡量各个降维方向上形成的灰度图像的质量。文中采用加权的Fisher判据来衡量图像的质量。在将图像分割成区域后,判据中的类间距离反映了区域之间的对比度,类内距离反映了区域内部之间的均匀性,权重反映了区域之间的相邻关系。这样将图像降维分成四步,先挑选样本图像,然后用混合高斯模型进行分割,再优化带权重的Fisher判据得到最优降维方向,最后利用最优降维方向将彩色图像转换成灰度图像。在对彩色IC样本图像进行降维的实验中,该方法能得到比其他方法质量更好的灰度图像。

In order to reduce the complexity of processing of color IC images, a color image is usually transformed into a gray one before further processing. The idea in data dimensionality reduction that the optimal projection direction can be obtained by optimizing a criterion was induced into color to gray transformation in this paper. In order to get the optimal direction, a criterion to evaluate the quality of a gray image obtained by different directions was needed. A new criterion called Weighted Fisher Criterion was proposed. After a color image was segmented into different regions, the new criterion evaluated the quality of an image by encouraging inner-region smoothness and inter-region contrast, especially the contrast between the neighboring regions which could be controlled by weights in the criterion. Thus the transformation from color to gray was divided into four steps, including selecting a color sample image firstly, segmenting it into regions based on Gaussian Mixture Model, and maximizing the Weighted Fisher Criterion to get an optimal direction to transform a color image into a gray one.