The detection of corrosion in steel sheets is an important task for the shipbuilding industry and is usually carried out through visual inspection. Some articles have analyzed this issue using digital image processing and have shown promising results. One of the stages of digital image processing involves the detection of edges, which can provide important and useful information about the corrosion state of the plate and characterizes the object's limits, being important for segmentation, registration or identification of objects. Many works have evaluated the images processed by the detection of edges with linear filters, some of them using the quality of the visual aspect of the processed image as a parameter, while others usually use entropy and PSNR. In this article, the objective of the work is to evaluate the performance of these filters and the behavior of entropy and PSNR applied to images of corroded plates. A corrosive solution was prepared in which six steel sheets were numbered and partially immersed in this solution. On each day, the plates were removed from the solution and photographed, the image being analyzed in the code. As a result, we conclude that entropy allows us to identify the best filter to analyze corrosion, while PSNR does not allow us to distinguish the difference in performance between the filters. Of the seven filters analyzed, the Canny filter showed the highest entropy value. On the other hand, the entropy values for each filter, apparently, do not depend on the corrosion time.

Keywords : Component; Formatting; Style; Styling; Insert (Key Words)