Segmentation and analysis of damages in composite images using multi-level threshold methods and geometrical features

• Published in: Measurement : journal of the International Measurement Confederation 2017-03, Vol.100, p.270-278
• Main Authors: Jac Fredo, A.R., Abilash, R.S., Kumar, C. Suresh
• Format: Article
• Language: English
• Subjects: Anisotropy; Composite materials; Damages; Diffusion; Digital images; Digital imaging; Eccentricity; Feature extraction; Image filters; Image segmentation; Impingement; Industrial applications; Mathematical analysis; Segmentation; Signal to noise ratio; Structural damage
• ISSN: 0263-2241; 1873-412X
• DOI: 10.1016/j.measurement.2017.01.002

[Display omitted] •Front and back view images of damaged composite materials are acquired.•Images are filtered, damaged regions are segmented and features are calculated.•Anisotropic filter gives better results compared to the median filter.•Tsallis multi threshold method is able to segment the damaged regions accurately.•Geometrical feature area is able to discriminate the 5mm, 6mm and 7mm damages. In this work, the digital images of damaged composite materials are analysed using multi-level threshold methods and geometrical features. The image are obtained from the front and back side of the composite material after applying 5mm, 6mm and 7mm impingement. Initially, the images are filtered using median and anisotropic diffusion filter. The results are validated using peak signal to noise ratio and structural similarity index measures. The damaged regions are segmented from the filtered images using threshold methods such as Otsu and Tsallis. The geometrical features such as area, perimeter, eccentricity and Major axis to Minor axis (MM) ratio are calculated from the delineated regions. The area calculated from the damaged region is correlated with the perimeter. The visual and validation results show that the median and anisotropic diffusion filter are able to remove the noise. The validation results suggest that anisotropic diffusion performs better than the median filter. The threshold methods are able to segment the damaged regions with the threshold level of four. Area and perimeter calculated from the delineated regions increases with increase in impingement. The damage dimension is seems to be high in the backside of the composite materials compared to the front side. The features calculated from the damaged region, extracted using Tsallis method is able to discriminate the damages better compared to the regions extracted using Otsu. The eccentricity of the damaged region increases and the MM ratio decreases with the increase in impingement. The shape of the damage extends from circle to ellipse and elevate towards the y-axis as the impingement increases. The area calculated from the damage region gives high correlation (R=0.99) with the perimeter. This suggests that the damage spreads on the composite material as a ring. The image based analysis carried out on this work is able to characterise the impairment in composite materials; this framework can be used for the industrial applications for the quantification of damages.