Captcha Recognition Based on Attention Mechanism

Captcha recognition is a worthful work to study, since it does help Internet security and also promotes the field of pattern recognition. In this work, we concentrate on the attention mechanism to this classification task for ameliorating the function of our baseline network. In our experiment, we arbitrarily combine the two modules in CBAM and the coordinate attention module that is considered to be efficient and novel. Then we add this combined attention to our baseline network. From the test results, we see that the better attention for this task is CBAM (spatial first, then channel attention), which improves the recognition accuracy to about 0.8% based on the mini-dataset generated by ourselves.