Handwritten Templated Sketch Recognition Based on a Hierarchical Method
Xin Yuan, Weiqun Cao · International Journal of Signal Processing Image Processing and Pattern Recognition · 2017
A hierarchical method for handwritten templated sketch recognition which combined bag of features (BOF) and perceptual hashing is proposed in this paper.The method takes both the overall properties and local characteristics of the sketch into account, in order to overcome the defection aroused by the strong randomness and much freedom of handwritten input.Firstly, we set up the rectangular bounding box for every sketch to get the corresponding sketch image and then adjust it to a square, the regularized sketch image.Secondly, divide every regularized sketch image uniformly to little patches, and take the bag of features as the local characteristics and use support vector machine (SVM) classifier to do the first level classification.Thirdly, resort Top 10 of the initial classification results using the perceptual hashing algorithm which reflects the differences of the objects on overall properties.For our experimental objects are sketches bearing certain structures (9 types), we take these structures as the additional overall features improving the recognition rate.We realize the recognition of 150 kinds of templated sketches, and the average recognition rate is 92.9% (Top1), and 100% (Top5) respectively.The experimental results show that the method is robust and has higher recognition rate.