Research on Image Recognition Based on Invariant Moment and SVM
Shi Jianfang, Bei Sun · 2010
Method of image recognition based on statistics can achieve fine performance only if large numbers of samples are provided. In some situation, it's impossible to obtain so many samples, which may result in the poor recognition-performance because lacking of information. Furthermore, frequently-used neural network is designed as classifier with the purpose of empirical risk minimization and with poor generalization. Consequently in the paper an arithmetic that combines wavelet moment with Support Vector Machine (SVM) is established to look for optimum solution of existing sample-information and is suitable for small sample analysis. In simulation, Hu, Zernike, and Wavelet moments of a finite number of tank images were extracted and recognized by BP neural net and SVM separately. Experimental results demonstrate that the arithmetic which combines Wavelet moments and SVM is superior to others on recognition efficiency in the case of small samples.