Fast semantic segmentation of aerial images based on color and texture

In this paper, a semantic segmentation method foraerial images is presented. Semantic segmentation allows thetask of segmentation and classification to be performedsimultaneously in a single efficient step. This algorithm relieson descriptors of color and texture. In the training phase, wefirst manually extract homogenous areas and label each areasemantically. Then color and texture descriptors for each areain the training image are computed. The pool of descriptorsand their semantic label are used to build two separateclassifiers for color and texture. We tested our algorithm byKNN classifier. To segment a new image, we over-segment itinto a number of superpixels. Then we compute texture andcolor descriptors for each superpixel and classify it based onthe trained classifier. This labels the superpixels semantically.Labeling all superpixels provides a segmentation map. Weused local binary pattern histogram fourier features and colorhistograms of RGB images as texture and color descriptorsrespectively. This algorithm is applied to a large set of aerialimages and is proved to have above 95% success rate.

کلیدواژه ها:

Semantic Recognition ، Texture descriptors aerial images ، superpixels