An efficient, scalable and robust approach to the handwritten digitsrecognition problem based on the Saak transform is proposed in this work.First, multi-stage Saak transforms are used to extract a family of jointspatial-spectral representations of input images. Then, the Saak coefficientsare used as features and fed into the SVM classifier for the classificationtask. In order to control the size of Saak coefficients, we adopt a lossy Saaktransform that uses the principal component analysis (PCA) to select a smallerset of transform kernels. The handwritten digits recognition problem is wellsolved by the convolutional neural network (CNN) such as the LeNet-5. Weconduct a comparative study on the performance of the LeNet-5 and theSaak-transform-based solutions in terms of scalability and robustness as wellas the efficiency of lossless and lossy Saak transforms under a comparableaccuracy level.
展开▼
