Geometrical feature extraction from ultrasonic time frequency images: An application to nondestructive testing of materials

Signal processing of ultrasonic scans is frequently used for material classification. Extracted parameters from the ultrasonic registers are used in statistical classification algorithms. There are situations where interactions taking place inside the material are so complex that the number of significative parameters is not enough to get good classification percentages. What we propose in this work is to use geometrical parameters extracted from the time frequency representation of the ultrasonic register in this situations. The presented technique merges some ideas of nondestructive testing of materials and image descriptors to get an easy to implement algorithm that could be applied in many different products and situations.