Electrical Inspection Oriented Thermal Image Quality Assessment

2016 International Conference on Communication, Image and Signal Processing

物理学;无线电电子学;计算机科学

Lin, Ying^1 ; Wang, Menglin^2 ; Gong, Xiaojin^2 ; Guo, Zhihong^1 ; Geng, Yujie^1 ; Bai, Demeng^1

State Grid Shandong Electric Power Research Institute, Jinan, China^1

Zhejiang University, Hangzhou, China^2

关键词: Human operator;    Information contents;    K-nearest neighbor method;    No-reference images;    Quality assessment;    Quantitative measurement;    Subjective assessments;    Two Dimensional (2 D);   

Others  :  https://iopscience.iop.org/article/10.1088/1742-6596/787/1/012009/pdf
DOI  :  10.1088/1742-6596/787/1/012009

学科分类：计算机科学（综合）

来源: IOP

【 摘 要 】

This paper presents an approach to access the quality of thermal images that are specially used in electrical inspection. In this application, no reference images are given for quality assessment. Therefore, we first analyze the characteristics for these thermal images. Then, four quantitative measurements, which are one-dimensional (1D) entropy, two-dimensional (2D) entropy, centrality, and No-Reference Structural Sharpness (NRSS), are investigated to measure the information content, the centrality for objects of interest, and the sharpness of images. Moreover, in order to provide a more intuitive measure for human operators, we assign each image with a discrete rate based on these quantitative measurements via the k-nearest neighbor (KNN) method. The proposed approach has been validated in a dataset composed of 2,336 images. Experiments show that our quality assessment results are consistent with subjective assessment.