The inspection process of metallic surfaces, especially FPSO tanks, is still heavily reliant on manual methods, requiring long production downtime and posing health risks to inspectors. Automating this analysis step will provide significant benefits to the management of these vessels' integrity, reducing expenses, downtime, and, most importantly, the exposure time of employees to hazards associated with inspection activities. During manual inspections, inspectors make annotations using paint, typically in white and yellow colors, directly on the tank walls, hindering the automation of the inspection process as it complicates the segmentation and identification of potential flaws on the tank wall using techniques such as neural network models. Recognizing this problem, this work presents a proposal for the identification and segmentation of these markings by segmenting them in the images, followed by the removal of the segmented markings using image texture-filling techniques.