Surface Quality Assessment with Advanced Texture Analysis Techniques Topi Mäenpää Machine Vision Group

Surface properties such as cloudiness and granularity can not be easily measured in a quantitative way. Traditional computer vision methods such as thresholding-based object feature detectors do not provide sufficiently reliable information. On the other hand, most texture analysis methods are overwhelmingly complex and slow. This paper presents a technique that uses statistical texture measures and a self-organizing feature map to automatically arrange textures into different categories. Unlike most other texture measures the proposed technique is simple and robust against artifacts such as illumination changes. Therefore, it performs better in many real world applications. Simplicity also means low computational burden which makes it possible to perform sophisticated analys in real time.