Efficient Convolutional Dictionary Learning When The Number Of Classes In The Dataset Increases

Convolutional sparse representation (CSR) is a method of representing an image by the convolution of a dictionary of image features and the coefficients representing the distribution of the dictionary. The convolutional dictionary learning algorithm using this CSR needs to be relearned from the beginning when the number of classes in the dataset to be trained is increased. In this research, we attempt to improve the learning speed when the number of classes in the dataset increases by using the dictionary learned on the dataset before the number of classes was increased.