Ziheng Sun, Liping Di, Hui Fang

Object based image analysis (OBIA) has a unique process requirement: relate all the pixels in the segmented images to the vectorized polygons (pixel in polygon). The existing solutions are very slow in finding the pixels in a polygon. This paper proposes a novel algorithm called Two-Pixel-Reference to speed up the process. The algorithm is initially designed for segmented remote sensing images. It avoids most multiple-layer loops in existing methods and trims many redundant comparison among pixels. Thus it has literal lower Big O algorithm complexity. We implemented the algorithms in C++ and made more than seventy tests on two different machines to compare the algorithm with three other existing algorithms. The results show that it significantly decreases the time cost of the process. In every single test, the proposed algorithm costs much less time than other algorithms. Specifically, the average duration is reduced from 3.96 seconds to 0.15 second on machine #1 and from 3.647 seconds to 0.073 second on machine #2. This paper makes a good example for researching time-efficient algorithms to accelerate the overall process of OBIA in such a big data era.