Image Fusion With Guided Image Filtering For Infrared And Visible Images Of Insulator

Insulator is one of the indispensable components in the substation equipment. With the development of smart grid, monitoring status of the substation equipment, especially insulator, is favored by researchers gradually. Insulator infrared image has the characteristics of the general infrared images, such as reflect hot target information, poor clarity, but typical feature. Insulator visible image can reflect scene information and edge information clearly, but not thermal information. Therefore, fusing insulator infrared image and visible image can improve the localization accuracy of heat faults. This paper mainly studies the image fusion algorithm of insulator infrared image and visible image,improves the existing fusion rules, completes the fusion experiments of insulator infrared image and visible image, and evaluate the fusion results from subjectively and objectively.As classical guided image filtering has the weakness of manually selecting parameters, this paper studies the adaptive selection guided image filtering added by similarity judgment, and completes the filter design with automatic selection of offset function and smoothing factor. Filter results show that compared with classical guide filtering, the proposed method has a great improvement in detail enhancement and edge preservation.Aiming at unique feature of insulator image, the fusion of insulator infrared image and visible image exists fuzzy edge information of umbrella plate, low luminance information and poor contrast. This paper mainly studies the image fusion method based on NSCT. Through the NSCT decomposition, the low-frequency subband mainly inherited energy information of the source image and the high frequency subbands contain edge information of the source image. Considering the rich edge information in high frequency subbands, and the edge-preseving function of classical guild filtering, this paper introduces classical guide filtering as high frequency fusion rules to NSCT, and low frequency uses energy of local area instead of single pixel as fusion rules.Experimental results show that the proposed method can improve the brightness of fusion results, and retain edge information of insulator umbrella plate clearly.The classical guide filtering in above methods does not meet adaptive selection of the parameter, and does not consider the probability correlation of energy of local area.Therefore, based on NSCT, this paper does the further research. High frequency uses the adaptive selection guided image filtering to realize edge-preseving and parameter adaptive selection. Low frequency uses the local correlation energy. By compared experiments, the fusion results of the proposed fusion method has low distortion, and this method can effectively combine the edge information of visible umbrella plate image and the heat information of infrared image.NSCT is belonging to multi-scale spare, low frequency has the weakness of sparse incomplete, and as regulation mathematical model, the training dictionary does not consider the characteristic between image. Joint sparsity model decomposes image into common parts and special parts by dictionary training. Therefore, this paper proposes the fusion method based on joint sparsity model and adaptive selection guided image filtering, and the fusion image processed by the proposed method has the high brightness and clear edge with favorable objective indicators.