Super-resolution for scene text images is a pre-processing of scene text recognition to improve recognition accuracy. This task aims to improve the visual quality of text regions in the images from low-resolution images. Although SR techniques have significantly improved with the recent development of deep learning, it is still challenging to reconstruct high-resolution images for wild images with irregular shapes, severe noise, and blurring. This is because CNN-based methods are based on local calculations and do not consider text-specific characteristics and these are unable to deal with irregular deformations, etc. In this paper, we propose a multi-task learning-based Attentional Feature Fusion Network (MAFF-Net) to reconstruct visually high-quality images from low-resolution images in real scenes. MAFF-Net consists of a reconstruction branch and a super-resolution branch, which are trained simultaneously to share complementary features of the reconstruction model, such as noise reduction and structural information of the text, using the feature representation transfer (FRT) module. In addition, the transformer module, equipped with a 2-D self-attention mechanism, is used to deal with irregular deformations of the text. Then, we attempt to improve the visual quality of the images with severe noise, blurring, and irregular deformations by fusing the attentional features of the different viewpoints of the FRT module and the transformer module, respectively. Experimental results on the benchmark TextZoom dataset show that the proposed method achieves competitive performance with state-of-the-art methods and proves its effectiveness, especially for challenging images.

In recent years, template-based methods such as Siamese network trackers and Correlation Filter (CF) based trackers have achieved state-of-the-art performance in several benchmarks. Recent Siamese network trackers use deep features extracted from convolutional neural networks to locate the target. However, the tracking performance of these trackers decreases when there are similar distractors to the object and the target object is deformed. On the other hand, correlation filter (CF)-based trackers that use handcrafted features (e.g., HOG features) to spatially locate the target. These two approaches have complementary characteristics due to differences in learning methods, features used, and the size of search regions. Also, we found that these trackers are complementary in terms of performance in benchmarking. Therefore, we propose the “Complementary Tracking framework using Average peak-to-correlation energy” (CTA). CTA is the generic object tracking framework that connects CF-trackers and Siamese-trackers in parallel and exploits the complementary features of these. In CTA, when a tracking failure of the Siamese tracker is detected using Average peak-to-correlation energy (APCE), which is an evaluation index of the response map matrix, the CF-trackers correct the output. In experimental on OTB100, CTA significantly improves the performance over the original tracker for several combinations of Siamese-trackers and CF-rackers.