Classifying for a mixture of object images and character patterns by using CNN pre-trained for large-scale object image dataset

摘要

Neural networks are a powerful means of classifying object images and character patterns. The proposed common classification method for object images and handwritten digits combines convolutional neural networks (CNNs) and support vector machines (SVMs). A pre-trained CNN, called Alex-Net, is used as a pattern-feature extractor. Alex-Net is pre-trained for the large-scale object-image dataset ImageNet. An SVM is used as trainable classifier. The feature vectors are passed to the SVM from Alex-Net. A mixture of STL-10 object images and MNIST handwritten digit patterns is trained by the SVM. Experimental test error rate for the mixture of test 8k STL-10 object images and 10k MNIST digit patterns was 7.734%, which shows that the proposed method is effective for common-category classification.