CT-Net: Cascade T-shape deep fusion networks for document binarization

• We present a Cascade T-Shape Deep Networks framework for document image binarization. The T-Net is multi-task enhancement and features learned by the are transferred to binarization which can improve performance of evaluated on nine public DIBCO datasets achieves new state-of-the-art performance. Document key step in most analysis tasks. However, historical-document images usually suffer from various degradations, making this very challenging processing stage. has improved dramatically recent years use Convolutional Neural (CNNs). In paper, dual-task, T-shaped neural network proposed that main task an auxiliary enhancement. learns degradations specific CNN-kernel be adapted towards training process. addition, considered as version input image, fed into fine-tuning, it possible design chained-cascade (CT-Net). Experimental results competition (DIBCO datasets) MCS dataset show our method outperforms competing methods cases.