<title>Comparison of image analysis techniques for defect detection in food-processing applications</title>

X-ray detection of defects and contaminants in the food industry

“…It is not practical to attempt to describe each method that has been presented in the literature. The majority of detection algorithms for X-ray images of food products use a variation of discriminant analysis [18], neural networks [19][20][21], or simple/adaptive thresholding [22]. Spectral filtering of the X-ray image has also been shown to be an effective tool [23,24].…”

“…Human recognition studies based on scanned film images have shown that X-ray imaging is extremely reliable for quality control purposes [57][58][59][60] but is obviously of no use for bulk inspection. While several effective automatic recognition algorithms have been reported [18,19,26,53,[61][62][63][64], in most cases the images used for training and testing were obtained using high resolution X-ray systems such as film, X-ray fluoroscopes, or X-ray microscopes that require exposure times that make bulk inspection unrealistic.…”

Computer vision for foreign body detection and removal in the food industry

“…The majority of detection algorithms for X-ray images of food products use a variation of discriminant analysis (Haff and Pearson, 2007), neural networks (Brosnan et al ., 1996;Casasent et al ., 1996;Talukder et al ., 1998), or simple/adaptive thresholding (Chen et al ., 2001b). Development of this component is nearly identical to the development of algorithms for optical inspection systems, so it will not be discussed in detail here.…”

“…X-ray system parameters for maximizing recognition of insects in wheat kernels have been investigated Schatzki and Keagy, 1991). While several effective automatic recognition algorithms have been reported (Brosnan et al ., 1996;Haff, 2001;Haff and Pearson, 2007;Jayas et al ., 2003;Karunakaran et al ., 2003aKarunakaran et al ., , 2003bKarunakaran et al ., , 2005Keagy and Schatzki, 1993), in most cases the images used for training and testing were obtained using high resolution X-ray systems, such as film, X-ray fluoroscopes, or X-ray microscopes that require exposure times, making bulk inspection unrealistic. While several effective automatic recognition algorithms have been reported (Brosnan et al ., 1996;Haff, 2001;Haff and Pearson, 2007;Jayas et al ., 2003;Karunakaran et al ., 2003aKarunakaran et al ., , 2003bKarunakaran et al ., , 2005Keagy and Schatzki, 1993), in most cases the images used for training and testing were obtained using high resolution X-ray systems, such as film, X-ray fluoroscopes, or X-ray microscopes that require exposure times, making bulk inspection unrealistic.…”