A Mixed Method for Feature Extraction Based on Resonance Filtering

Machine learning tasks such as image classification need to select the features that can describe well. The has individual and common features, they are interdependent. If only of emphasized, neural network is prone overfitting. images networks will not be able adapt diversified environments. In order better integrate based on skeleton edge extraction, this paper designed a mixed feature extraction method resonance filtering, named layer. Resonance layer in front input layer, using K3M algorithm extract skeleton, Canny border, filtering reconstruct training by noise, through set efficient expression characteristics improve efficiency network, so accuracy prediction. Taking fully connected LeNet-5 for example, experiment handwritten digits database shows proposed while out part noise data.