Identifying defects under shadows from a series of thermal images by employing robust principal component analysis and local binary function

Abstract

Thermal infrared images have been widely employed to detect defections. However, it is challenging to identify the defects from thermal infrared images covered with shadows and noise. For a series of thermal infrared images, principal component analysis (PCA) is often applied to transform the given series into a lower-dimensional linear subspace. In the lower-dimensional subspace, a low-rank matrix is generated to serve as an optimal estimation from the series of thermal images. However, the information stored in those thermal infrared images will be kept mostly during the transformation. Full recovery of the lost information is not possible. A template containing the major information from the given series can be extracted by employing PCA. Furthermore, PCA has difficulty finding the template should interferences occur while the thermal images are captured. The unnecessary information collected associated with the interferences causes some unfavorable characteristics of the template extracted by PCA. Robust PCA (RPCA) is less susceptible to the abovementioned constraints. In this study, RPCA is employed to extract a template from a series of thermal infrared images. Local binary functions are built to restore the image free of noise by keeping the local boundaries. The defects can be readily identified from the regional boundaries. The proposed approach combining RPCA and local binary functions to analyze the given images in conjunction with level set functions. The processed results demonstrate that the proposed scheme are more effective than PCA in analyzing a series of thermal infrared images containing interferences.