Analysis of Recognition Performance of SVMs Based on Three Types of Common Feature Datasets

摘要

The recognition performance of a support vector machine (SVM) is seriously affected by the feature datasets it trains. SVMs with different type of feature datasets have different recognition performances. Three type of common feature datasets, including gray value dataset, Hu moment dataset, and principal component dataset, are studied by experiments in this paper; results show that SVMs based on the gray value dataset is applicable to the low-resolution image recognition; SVMs based on the Hu moment dataset is applicable to the high-resolution image recognition; SVMs based on the principal components dataset is too time-consuming to the real-time image recognition.