Development of Digital Inspection Algorithms for X-Ray Radiography Casting Images

Abstract

This manuscript is concerned with the development of digital inspection of casting defects using x-ray radiography images. An efficient approach for detection and classification time of casting defects in x-ray radiography images is proposed. The accuracy of this approach depends on suggested algorithms for background correction, image de-noising, image enhancement and image segmentation of casting defects. Three different algorithms are introduced for automatic detection of casting defects in x-ray images. These algorithms depend on features extraction power density spectrum (PDS) and high order statistics (HOS). An artificial neural network is utilized as a classifier for matching purposes of extracted features. The results show that HOS achieved the best recognition rate of 100% for casting defects in X-ray radiography images in comparison with other algorithms. Besides, a reduction of classification time for casting defects is another target in this paper. It is achieved using costly powerful digital processing hardware and advanced software. Furthermore, an algorithm is realized to reduce classification time of casting defects. This algorithm depends on textural features that extracted from x-ray images of casting defects. Hence, a feature reduction program code is implemented for reduction of extracted features. This program code is relied on average value of each extracted feature for normal and defect image. The numbers of extracted features are reduced from 22 to 2 features. Therefore, better execution time can be achieved for classification purposes of casting defects. The proposed algorithms are evaluated using Intel core TM i5-3470 CPU with 3.20 GHz and Intel core TM i7-3612QM CPU with 4.00 GHz. Consequently, these algorithms can be transferred into more powerful digital processing hardware such as FPGA and GPU for faster classification of casting defects. The obtained results confirm that proposed algorithms can be applied for a broad range of non-destructive applications using image processing techniques.