| Compared with visible imaging, infrared iamging has the advantages of strongpenetration, good concealment, and little impact of weather conditions,etc, so it is widelyused in military area and in civilian area. Howerer, Contrast to visible imaging,there are manyshortcomings to infrared image also, such as low signal noise ratio, low contrast ratio, etc.Therefore, automatic target recognition in infrared image processing is more difficult, and it isalso becoming a research hotspot.Automatic target recognition in infrared image is a complex process, which involvesalmost all the aspects of image processing. The main research work that has done in this paperis as follows:1. Several commonly used denoising methods were introduced briefly aiming at solvingnoise problems in the infrared image. A new ROI detection method was proposed aiming atsolving interest detection problem which exists in information entropy method at the stage ofimage segmentation. First, the original image is divided into super pixels, and targets in theimage were fully divided into different super pixels. Secondly, the variance weightedinformation entropy was calculated for each of the super pixels, then the interest region wasdetected which contains the targets. Finally, the precise segmentation of the image was made.2. Several commonly used feature extraction methods are introduced, and the histogromof oriented gradient (HOG) feature of object was deeply studied at the stage of target featureextraction in infrared image. In order to solve the large dimensions problem of HOG features,a hierarchical HOG feature extraction method was proposed which can greatly reduce thefeature dimensions and the calculation amount of the following target recognition.3. At last, some feature classification methods are introduced briefly, and the system isimplemented by the method of support vector machine. The experimental results wereanalysised, and compared to the existing scheme, the effectiveness of the method is tested. |
PDF Full Text Request