Multi-Feature Fusion Aerial Image Segmentation in Complex Background

The hybrid method of initial partitioning and region merging is widely used in aerial image segmentation. The existing initial partitioning methods are the watershed transform of the Edge Strength Map (ESM) of aerial images. Therefore, if watershed algorithm is used in images with discontinuous edges and lots of noise, it will be easy to produce "improper segmentation". In order to form high-quality initial partitions, we propose a new MRF (YMRF) image segmentation method from the perspective of fully exploiting the image spatial information. The key points of region merging are region similarity measurement, merging process and merging stopping moment, but the problem of region label selection is ignored after the region pair which will be merged is selected. So, we propose a kind of image scene to reflect the necessity of paying attention to this problem and develop a region merging label selection mechanism for the image scene. To solve the problem that merging stopping moment tends to form the result with high homogeneity in the domain, we propose a optimal merging state, which can weaken the homogeneity in the domain. Experimental results show that our algorithm is more effective than the existing methods, when they are used in our unique dataset.