A New Convolution Network Based on Laplacian Eigenmap

The best is to read these instructions and follow the outline of this text. Recently, convolutional deep neural network (ConvNet) has been widely used in the field of image classification. In this work, we propose a new feedback free convolution network for image classification. The proposed network could hierarchically and effectively extract the features from an image through a manually designed convolution network without relying on back-propagation. The network is designed in a cascaded fashion, where the Laplacian Eigenmap filter is used as convolution kernel to extract features in each of the cascaded stage. The final output of the network is achieved by a simply binary hashing and histogram encoding, and could be served as distinguishing features for many classification tasks. Experiments on different database, e.g. FERET datasets for face recognition, CUReT for texture classification and MNIST for hand-written digits recognition, showed that the proposed method outperforms many other popular machine learning algorithms.