In this paper we present a computer vision approach developed for defects detection and wall thickness measurements of corroded pipes inspected by tangential radiography. The main objective of the work is to optimized data collection and film interpretation using computerized image analysis. The proposed methodology can be resumed as follow. First, a gradient operator is used to locate the pixels which defines the frontier between the pipe and the film background. Secondly, an adapted regression is performed on the pixels to fit the pipe outer contour. The inner contour is then located based on the luminance maximum of segments oriented perpendicularly to the fit. From there, wall thickness measurements can simply be obtained by subtracting one contour to the other. Another important system feature is its adaptability to pipe wall texture. In other words, image analysis is refined in potentially defective or corroded areas. This allows precise thickness measurements and shape identification in critical segments. Finally, the system performance is tested by analysing different pipe geometries such as straight segments, elbows, T and welded junctions. The studied samples are radiographs gathered from industrial applications which covers a wide variety of corrosion profile.