Morphological Modified Global Thresholding and 8 Adjacent Neighborhood Labeling for SWIR Image Mosaic

Abstract

Due to the proper structure and the nonuniformity of the imager detectors, the mosaic algorithms for SWIR (Short-Wave InfRared) remote sensing images slightly differentiate from common mosaic techniques. A detailed description of a novel mosaic method has been proposed in the paper. Firstly, pixels re-arranging and nonuniformity verifying have been done with the images during the pre-processing step. Scene-based statistical method is applied to remove most of the nonuniformity noise. Secondly, a novel feature-based image registration algorithm is studied in the registration step. The morphological modified global thresholding algorithm is proposed to segment the two images. Then region features are extracted by 8- adjacent neighborhood labeling method. The area of the regions and the slope between their central points are used in feature matching. Thirdly, the affine transformation between the images is determined by the matched features and the information from the two images is fused to make a larger format image by the weighted-averaging method. At last, two pairs of images after pre-processing are used for registration experiment while a pair of images is registered before noise removing. The registration results are tested to be insensitive to noise and be excellent and reliable even though the images are slightly twisted. The fusion experiment is done for non-twisted images and the mosaic result shows that our method is efficient in the procedure and can meet the requirement of remote sensing image process.