Energy Efficient Text Spotting Technique for Mobile Edge Computing

We propose a new scene text spotting system which aims at minimizing the power consumption in mobile edge devices. We focused on preprocessing methods and changes in the processing pipeline. Overall, we removed non-text images from the processing pipeline to reduce power consumption and positioned texts at the center of the images to improve accuracy. Moreover, we were able to achieve a substantial power saving and increased text recognition accuracy. Compared to the baseline method, our proposed method shows an 80% higher performance score. The accuracy score was increased by 17% and the power consumption was reduced by 30% because we could reduce the execution count of the neural network by 40%.