Applied dual Bi-Directional LSTM for image classification through max-pooling the forward and backward LSTM’s hidden states of both the image matrix and its transpose and concatenating them for inputting into a dense fully connected layer for classification.