Text Gestalt: Stroke-Aware Scene Text Image Super-Resolution

Abstract

In the last decade, the blossom of deep learning has witnessed the rapid development of scene text recognition. However, the recognition of low-resolution scene text images remains a challenge. Even though some super-resolution methods have been proposed to tackle this problem, they usually treat text images as general images while ignoring the fact that the visual quality of strokes (the atomic unit of text) plays an essential role for text recognition. According to Gestalt Psychology, humans are capable of composing parts of details into the most similar objects guided by prior knowledge. Likewise, when humans observe a low-resolution text image, they will inherently use partial stroke-level details to recover the appearance of holistic characters. Inspired by Gestalt Psychology, we put forward a Stroke-Aware Scene Text Image Super-Resolution method containing a Stroke-Focused Module (SFM) to concentrate on stroke-level internal structures of characters in text images. Specifically, we attempt to design rules for decomposing English characters and digits at strokelevel, then pre-train a text recognizer to provide stroke-level attention maps as positional clues with the purpose of controlling the consistency between the generated super-resolution image and high-resolution ground truth. The extensive experimental results validate that the proposed method can indeed generate more distinguishable images on TextZoom and manually constructed Chinese character dataset Degraded-IC13. Furthermore, since the proposed SFM is only used to provide stroke-level guidance when training, it will not bring any time overhead during the test phase. Code is available at https://github.com/FudanVI/FudanOCR/text-gestalt.

Learning-Based Text Image Quality Assessment with Texture Feature and Embedding Robustness

“…In addition, more fine-grained text detail constraints are proposed. Such as character skeleton loss [10] and stroke-focused module [11].…”

“…Text image quality has gradually become an important factor for highperformance STR systems. Recent works improve the recognition performance on superresolution (SR) dataset Textzoom [6] and benchmarks by introducing SR methods as a pre-processing procedure before recognition [6,[8][9][10][11] or joint training [4,12].…”

“…The main influencing factors of text image quality include image properties (such as low-resolution (LR), blurriness, contrast, and background) and text properties (such as sloped, curved, and irregular shapes). This is why character position [8], character skeleton [10], and character stoke [11] are essential for recovering text images with more details.…”