Superpixel Regions Extraction for Target Detection

In this paper, an algorithm of target region detection is proposed based on superpixel segmentation in the field of computer vision which is imported to high-resolution remote sensing images for superpixel-level rather than pixel-level target detection. For the problem of massive data, redundant information and time-consuming targets searching of high-resolution remote sensing images with complex scene and large size, the region of interest (ROI) extraction strategy based on a visual saliency map detection is adopted. Second, the multidimensional description vector of local feature is constructed via superpixels obtained from simple linear iterative clustering (SLIC). Third, combine with the prior information of the target to determine the threshold of feature, from which we select the candidate superpixels belong to target. Experimental results show that the proposed algorithm is more effective in high-resolution remote sensing images, overcoming the situation of complex background interference and robust to the target rotation. In addition, the proposed algorithm performs favorably against the traditional sliding windows search algorithm. On one hand, significantly reduces the computing complexity of the search space, and achieves the data dimensionality reduction. On the other hand, it brings lower false probability and improves the detection accuracy.