One DAG to Rule Them All

In this paper, we present novel strategies for optimizing the performance of many binary image processing algorithms. These are collected in an open-source framework, graphgen , that is able to automatically generate optimized C++ source code implementing desired optimizations. Simply starting from a set rules, algorithms introduced with framework can decision trees minimum average path-length, possibly considering pattern frequencies, apply state prediction and compression by use directed rooted acyclic graphs (DRAGs). Moreover, proposed algorithmic solutions allow combine different optimization techniques significantly improve performance. Our proposal showcased on three classical widely employed (namely Connected Components Labeling, Thinning, Contour Tracing). When compared existing approaches —in 2D 3D—, implementations using generated optimal DRAGs perform better than previous state-of-the-art algorithms, both CPU GPU.