Image-free multi-character recognition

The recently developed image-free sensing technique decouples semantic information directly from compressed measurements without image reconstruction, which maintains the advantages of both the light hardware and software. However, the existing attempts have failed to classify multi-semantic information with multiple targets in the practical fieldof-view. In this Letter, we report a novel image-free sensing technique to tackle the multi-target recognition challenge for the first time, to the best of our knowledge. Different from the convolutional layer stack of image-free single-pixel networks, the reported convolutional recurrent neural network (CRNN) uses the bidirectional LSTM architecture to predict the distribution of multiple characters simultaneously. The framework enables capture of the long-range dependencies, providing a high recognition accuracy of multiple characters. We demonstrate the technique's effectiveness in license plate detection, which achieves a recognition accuracy of 87.60% at a sampling rate of 5% with a refresh rate higher than 100 FPS.