Research On Adaptive Stripe Non-uniformity Correction Algorithm For Infrared Image

In recent years,with the steady development and maturity of infrared thermal imaging technology,infrared imaging systems have become more and more widely used in civil and military fields.However,limited by the existing process level and manufacturing cost,stripe non-uniformity is common in the original infrared image,which has an adverse effect on the visual effect of the image and subsequent detail and contrast enhancement,as well as object detection and recognition.And it is often difficult to distinguish it from the edge in the image in the correction process.Therefore,it is necessary to study a correction algorithm that can make a balance between effectively eliminating stripe non-uniformity and protecting image edges.At present,a new stripe non-uniformity correction algorithm based on local spatial correlation can eliminate fringe non-uniformity effectively and it takes edge protection into account.Aiming at the problem that the algorithm still needs to be improved,especially the fixed detection threshold in edge detection,which makes the algorithm unable to adapt to the stripe non-uniformity correction of different types of images collected by different infrared detectors,an improved algorithm is proposed in this paper.Firstly,from the view of adaptively selecting edge detection threshold according to image characteristics,an edge detection method which adaptively chooses edge detection threshold according to the statistical characteristics of absolute value of image horizontal gradient is proposed.The minimum horizontal gradient absolute value corresponding to 99% of the total number of pixels in the statistical histogram of the original image is taken as the edge detection threshold,and then whether the pixel is in the edge region of the image is judged in the detection window of 1*3 size.Secondly,based on this study,the method of calculating the expected value of the pixel judged by the threshold as the non-edge area of the image is optimized.Instead of using the average value of all the pixels in the 1*3 window as the expected output value,the average value of the left and right adjacent pixels is used as the expected value according to the similarity of the gray value of the pixels in the horizontal local space,so as to further improve the real-time performance of the algorithm.The proposed algorithm together with several other typical algorithms were compared from subjective and objective perspectives on real infrared images collected by different detectors,including different types of scenes.Experimental results show that the proposed algorithm has good scene adaptability compared with the existing algorithms.It can be applied to the correction of different types of scenes collected by different detectors.It can effectively remove the non-uniformity of fringes and better protect image edges,and it further improves the real-time performance of the previous algorithm..