Fast semantic segmentation of aerial images based on color and texture

In this paper, a semantic segmentation method for aerial images is presented. Semantic segmentation allows the task of segmentation and classification to be performed simultaneously in a single efficient step. This algorithm relies on descriptors of color and texture. In the training phase, we first manually extract homogenous areas and label each area semantically. Then color and texture descriptors for each area in the training image are computed. The pool of descriptors and their semantic label are used to build two separate classifiers for color and texture. We tested our algorithm by KNN classifier. To segment a new image, we over-segment it into a number of superpixels. Then we compute texture and color descriptors for each superpixel and classify it based on the trained classifier. This labels the superpixels semantically. Labeling all superpixels provides a segmentation map. We used local binary pattern histogram fourier features and color histograms of RGB images as texture and color descriptors respectively. This algorithm is applied to a large set of aerial images and is proved to have above 95% success rate.