Image Augmentation for Object Image Classification Based On Combination of PreTrained CNN and SVM

Neural networks are a powerful means of classifying object images. The proposedimage category classification method for object images combines convolutional neuralnetworks (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-imagedataset ImageNet. Instead of training, Alex-Net, pre-trained for ImageNet is used. An SVM isused as trainable classifier. The feature vectors are passed to the SVM from Alex-Net. TheSTL-10 dataset are used as object images. The number of classes is ten. Training and testsamples are clearly split. STL-10 object images are trained by the SVM with dataaugmentation. We use the pattern transformation method with the cosine function. We alsoapply some augmentation method such as rotation, skewing and elastic distortion. By using thecosine function, the original patterns were left-justified, right-justified, top-justified, or bottomjustified. Patterns were also center-justified and enlarged. Test error rate is decreased by 0.435percentage points from 16.055% by augmentation with cosine transformation. Error rates areincreased by other augmentation method such as rotation, skewing and elastic distortion,compared without augmentation . Number of augmented data is 30 times that of the originalSTL-10 5K training samples. Experimental test error rate for the test 8k STL-10 object imageswas 15.620%, which shows that image augmentation is effective for image categoryclassification.