Detection of low-contrast objects in textured images

摘要

In this paper we present a method for the detection of objects that are not clearly defined by an edge within the underlying texture. The application of this method is to detect impurities or contaminants within food products. The authors have previously proposed a system that has an extremely high detection rate for a wide range of contaminants, but which needs to be further developed for the detection of low contrast contaminants. The method presented in this paper uses convolution to extract texture features from the food images to generate the texture energy images. The convolution mask coefficients are the principal components obtained from images that do not have any foreign objects. The grey levels of the resulting texture energy images are modified to eliminate the underlying noise in a consistent way across all these images. A distance map image is created using the Mahanalobis distance measure to indicate the presence of any contaminants within the food products. This paper shows that the proposed method can cope with the subtle variations between the contaminants and the food background and successfully detect the low contrast contaminants.