Texture feature coding method for SAR automatic target recognition with adaptive boosting

In this paper, the applicability of texture feature coding method (TFCM) to synthetic aperture radar (SAR) automatic target recognition (ATR) is studied. The TFCM cooccurrence matrix (CM) is used as an additional image feature for adaptive boosting (AdaBoost) algorithm which uses originally 2D-DFT coefficients as an image feature. The TFCM CM extracts the connected texture information from the image while 2D-DFT gives the value of the spatial frequency components. The TFCM CM is invariant under the rotation of images. With these characteristics, the TFCM CM can discriminate a confused target which is not classified properly using 2D-DFT. The TFCM CM is combined with 2D-DFT in fusion process of the AdaBoost algorithm. In experimental results, it is shown that the correct-classification probability of the proposed scheme is larger than that of the conventional scheme which uses only 2D-DFT and raw image as image features.