A Method for Correction and Enhancement of Non-Uniformly Illuminated Images

Non-contact imaging detection of explosives and other hazardous materials is currently a hot spot for research in the field of explosives detection technology, and near-infrared hyperspectral detection technology based on multivariate optical calculations is one of the feasible means. However, in practical applications, the application of this technology is limited due to uneven illumination, which is prone to false alarms and missed alarms. In this paper, we first briefly introduce the explosive imaging detection system based on multivariate optical computation and analyze the characteristics of the scored image; then we adopt the multi-scale top-hat transformation to calculate the background illumination and correct the light inhomogeneity of the scored image; finally, we convert the scored image into a signal-to-noise ratio image through the steps of image segmentation and calculation of signal-to-noise ratio, which further solves the problem of threshold drift caused by light inhomogeneity and thus greatly improves the accuracy of recognition results is greatly improved. The experimental results show that the recognition accuracy and stability of the method in this paper are higher compared with the previous recognition results, which provides technical support for practical applications.