An innovative image-processing model for rust detection using Perlin Noise to simulate oxide textures

• A modeling study for simulation and detection of rust in images is proposed.
• Rust is synthesized using Perlin Noise in order to analyze extreme rust conditions.
• Higher persistence levels for Perling Noise yield higher classification errors.
• Only the mean was extracted to detect rust, lowering the computational load.
• The model’s performance is fast and adequate for industrial settings.

This article presents an image-processing model for detection of rust zones using digital images of metals. The input image, containing a wide range of possible rusted textures, is simulated with Perlin Noise, which allows simulating extreme corrosion conditions, without waiting for these conditions to occur. Probabilistic descriptors are determined by means of discriminant analysis using Fisher indexes. A Bayesian classifier is used to identify rusted regions. Additionally, performance tests under different noise conditions and texture variations, generated with Perlin Noise, are presented.

Figure optionsDownload as PowerPoint slide