Infrared Plane Image Recognition Based On Image Features Matching

As the character of the infrared imaging, it’s very difficult to detection and recognition ofthe infrared image: firstly, the distance between the target and the image is very long, the pixel ofthe target in the image is very little; secondly, for the disturb of the noise and the wave of thebackground the signal of the target is very weak, even flood in the strong noise; finely, as there isonly a little useful texture features, the information for detection and recognition is very finite.Therefore, the infrared target detection and recognition for in-depth study has importanttheoretical significance and practical value.Paper focuses on the research of different models of plan’s infrared image detection andrecognition:first, due to its characteristics of infrared image, it makes the segmentation of infraredimage become very difficult, through analyzed and compared the maximum entropysegmentation method, fuzzy threshold segmentation method and the adaptive thresholdsegmentation method, it can be seen that adaptive threshold method suitable for the segmentationof infrared image. Based on the research object characteristic and adaptive threshold method ofimage segmentation results, select the edge detection method, through compared the Robertsoperator, Soble operator and Canny operator, select the Roberts edge detection operator forprocessing.Secondly, studied on the principle of infrared image, aiming at the relationship betweeninfrared target and background, select of infrared characteristic quantity, ratio of length to width,complexity, compactness, part of the brightest pixel points and the target pixels total ratio,average contrast, form the feature vector for preliminary recognition of infrared image.Finally, in the process of the preliminary identification, the images has been simplified, thenusing the SIFT algorithm for accurately identify, mainly to exclude the error identification in the rough recognition, as the images was simplified, the recognition time is obviously shorten thanusing SIFT algorithm directly.Through the simulation analysis of the experiment results shows that, the thesis using theinfrared feature combine with SIFT algorithms for image recognition, shorten the recognitiontime of0.06s, recognition accuracy is improved greatly, achieve98%above.