Sketch Recognition Based On Graph Edit Distance

Sketch recognition is a new branch of face recognition, and it searches for the corresponding photos of the sketch in the existing-photo database. It widely applied in commercial and law area, such as public security, case-solving, suspect-searching and so on. So we do lots of work to research sketch recognition, and propose some new ideas.A method of sketch-photo recognition based on Graph Edit Distance of Edge Direction Histogram is proposed. The method uses the graphs to describe the feature of sketches and photos, and obtains GED of the graphs by combining EDH and Earth Mover's Distance. GED is used as the measurement of the similarity, and we achieve sketch recognition through nearest neighbor recognition methods. Experiments show that this method is effective in sketch-photo recognition.Comparing with the above-mentioned methods, a new method based on the Graph Edit Distance of Hidden Markov Model is proposed. After expressing the sketches and the photos with graphs, we calculate the similarity between the sketches and the photos using GED, and GED combines HMM and Kullback-Leibler Distance. So we can attain the sketch-photo recognition. In the premise signing the nodes most correctly, this method is mostly effective of high accuracy in sketch-photo recognition.Finally, the ways motioned above build the graphs of the sketches and the photos based on the premise of signing nodes. The method isn't only cost time, heavy workload, but also very difficult to increase the accuracy rate. So we propose a new method of sketch recognition based on Active Appearance Model and GED. First we build Active Appearance Model to do a search among the sketches and photos, and we can attain the final optimized nodes of the sketches and photos. In use of the final optimized nodes, we distinguish sketches from photos combining with GED. Experiments show that, this method greatly reduces the man-made factors on sketch recognition, and has better versatility and maneuverability in sketch recognition, and the performance is greatly improved.