Evaluation of grinding surface roughness based on gradient similarity and color similarity

Existing machine vision detection methods for surface roughness based on design indices are mainly divided into two categories: grayscale information-based and color information-based. The former loses a large amount of image information when converting the image's three-dimensional color space to one-dimensional grayscale space. The latter does not fully consider the color information and structural detail changes in images of different rough surfaces. To address the above problems, a visual measurement method of grinding surface roughness based on gradient similarity and color similarity is proposed in this study. This method purposefully uses gradient similarity and color similarity to evaluate the structural differences and color differences between different roughness images, respectively. The comparison experiments with CD, F2, and F5 indices show that the GC index we proposed has a strong correlation with the grinding surface roughness, and its regression fitting prediction model has a high prediction accuracy. In addition, we have discussed the effect of light intensity on the GC index. The experimental result indicates that the correlation between the GC index and roughness is relatively stable under different light intensities.