Logo Recognition Based On Visual Saliency

There are many kinds of daily goods.It is very important for buyers to make shopping decisions if they can get the information of goods easily and quickly.In order to solve this problem,the method of logo image recognition is studied,and a logo recognition system is designed and implemented.The main work is as follows.First,a segmentation method based on visual saliency is proposed.Based on the RC algorithm,an improved method MLRC,is presented.Two improvements are made in MLRC.The one is to change the single-scale RC method to multi-scale one,the other is to further fuse multi-scale RC saliency maps with multiple location priors maps.The saliency maps obtained by the algorithm further eliminate the background redundancy information.Experimental results show that most of the logo images can be segmented correctly when the saliency maps obtained by the MLRC algorithm are processed by threshold segmentation.Second,a multiple feature selection method based on improved genetic algorithm is proposed and a logo classifier based on SVM is constructed.In order to improve the recognition accuracy,multiple types of features are extracted to enhance the representation and differentiation of features.In order to improve the recognition efficiency,the genetic algorithm is used to select multiple features.In order to improve the validity of the fitness function,the k-fold cross validation method based on SVM is used to calculate the classification accuracy.The results show that the optimal feature subset obtained by the algorithm not only reduces 81 feature dimensions,but also improves the accuracy of the SVM classifier by 1.29%.Third,a logo recognition system is designed and implemented.The system contains logo image acquisition module,image processing module,feature processing module,recognition module and so on.The system is implemented by C++ and Matlab.