Content and Style Aware Generation of Text-Line Images for Handwriting Recognition

Abstract

Handwritten Text Recognition has achieved an impressive performance in public benchmarks. However, due to the high inter-and intra-class variability between handwriting styles, such recognizers need to be trained using huge volumes of manually labeled training data. To alleviate this labor-consuming problem, synthetic data produced with TrueType fonts has been often used in the training loop to gain volume and augment the handwriting style variability. However, there is a significant style bias between synthetic and real data which hinders the improvement of recognition performance. To deal with such limitations, we propose a generative method for handwritten text-line images, which is conditioned on both visual appearance and textual content. Our method is able to produce long text-line samples with diverse handwriting styles. Once properly trained, our method can also be adapted to new target data by only accessing unlabeled text-line images to mimic handwritten styles and produce images with any textual content. Extensive experiments have been done on making use of the generated samples to boost Handwritten Text Recognition performance. Both qualitative and quantitative results demonstrate that the proposed approach outperforms the current state of the art.

Detecting Tool Keypoints with Synthetic Training Data

“…To evaluate their readability, the Character Error Rate (CER) is also adopted. Efforts towards the definition of a task-specific score for HTG have recently brought to the vFID score [8] and the Handwriting Distance (HWD) score [54]. In this work, we consider a number of evaluation scores and settings to favor the adoption of our proposed evaluation protocol and to provide a meaningful comparison between existing HTG approaches.…”

APRNet: Attention-based Pixel-wise Rendering Network for Photo-Realistic Text Image Generation

“…Afterwards, the cGAN [34] allows controlling the class of generated contents by inputting an additional class condition. The latest techniques [3], [35]- [40] focus on line-level style-guided text image generation. These systems synthesized glyph-realistic text line images with different writer styles, usually optimized by discriminators, writer classifiers, and character recognizers.…”

“…S TYLE-GUIDED text image generation is a challenging task, which tries to synthesize text images by imitating reference style image's appearance while keeping text content unaltered (e.g. [1]- [3]). The appearance of a text image includes foreground and background color patterns, spatial transformations and deformations, typography (for printed text), writing styles and stroke thickness (for handwriting text), etc [2].…”

“…Recently, neural network based style-guided approaches have been applied to printed [7], [8] and handwriting text image generation [1]- [3]. Equipped by the generative adversarial network (GAN) [9], style transfer [10]- [15] and imageto-image translation [16]- [19], the quality of synthetic text images has been improved greatly.…”

Development of Language Models for Continuous Uzbek Speech Recognition System

“…Continuous speech recognition requires an LM to be built into an efficient system. An LM is essential for many applications in natural language processing (NLP), including but not limited to handwriting recognition [ 1 , 2 , 3 , 4 ], machine translation [ 5 ], speech recognition [ 6 , 7 ], integral model [ 8 ], phonemes [ 9 ], and vowels [ 10 ].…”