Top-down and bottom-up cues for scene text recognition

By Anand Mishra, Alahari Karteek, C. V. Jawahar. 2012 IEEE Conference on Computer Vision and Pattern Recognition, 2012

357 citations recorded by the source corpus.

Subfields: Computer Vision and Pattern Recognition, Media Technology

View Paper: https://www.semanticscholar.org/paper/66b71064b99331f908b60cb6d138f2ebea5bdcca

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Computer Vision and Pattern Recognition

1. Xiang Bai
2. Luc Van Gool
3. Andrew Zisserman
4. Andrea Vedaldi
5. Chunhua Shen
6. Dacheng Tao
7. Yu Qiao
8. Cong Yao

In Media Technology

1. Cong Yao
2. Xiang Bai
3. Minghui Liao
4. Xin He
5. Yu Qiao
6. Chunhua Shen
7. Weilin Huang
8. Albert Gordo

Citations in the corpus

• Show, Attend and Read: A Simple and Strong Baseline for Irregular Text Recognition (2018, background): “IIIT 5K-Words (IIIT5K) (Mishra, Alahari, and Jawahar 2012a) contains 5000 word patches cropped from natural scene images found by Google image search, 2000 for training and 3000 for test.”
• A robust arbitrary text detection system for natural scene images (2014, background): “Despite many efforts (Mishra et al., 2012; Smith et al., 2011; Wang et al., 2011; Wei & Lin, 2012; Weinman, Learned-Miller, & Hanson, 2009) for text detection from natural scene images, it is still considered a challenging and unsolved problem.”
• PhotoOCR: Reading Text in Uncontrolled Conditions (2013, methodology): “The method of [15] also relies on small lexicons to allow for a strong bigram constraint in a conditional random field (CRF) model.”
• LVLM-EHub: A Comprehensive Evaluation Benchmark for Large Vision-Language Models (2023, neutral): “We hope that our conclusions can pave the way for developing innovative strategies aimed at enhancing multimodal techniques.”
• Label embedding for text recognition (2013, methodology): “Experiments are performed on two challenging datasets (one of license plates and one of scene text) and show that the proposed method is competitive with standard bottom-up approaches to text recognition.”
• Symmetry-Constrained Rectification Network for Scene Text Recognition (2019, methodology): “A popular pipeline of traditional methods [10, 50, 49, 51, 38, 37, 39, 55, 4] is in the bottom-up architecture.”
• Image Retrieval Using Textual Cues (2013, methodology): “content in the image into text, either with [15] directly, or by localizing with [9], and then recognizing with [13].”
• TrOCR: Transformer-based Optical Character Recognition with Pre-trained Models (2021, methodology): “Here we leverage some widely-used benchmarks, including IIIT5K-3000 (Mishra, Alahari, and Jawahar 2012), SVT-647 (Wang, Babenko, and”
• Scene text detection and recognition: recent advances and future trends (2015, methodology): “[31], since it also used cues from both low level and high level”
• Supervised mid-level features for word image representation (2014, methodology): “The most popular approaches to address this task involve detecting and localizing individual characters in the word image and using that information to infer the contents of the word, using for example conditional random fields and language priors [39, 19, 20, 5, 22].”
• OCRBench: on the hidden mystery of OCR in large multimodal models (2023, positive): “Text Recognition : We evaluate LMM using widely-adopted OCR text recognition datasets, including (1) Regular Text Recognition: IIIT5K [37], SVT [38], IC13 [39]; (2)Irregular Text Recognition: IC15 [40], SVTP [41], CT80 [42], COCOText(COCO) [43], SCUT-CTW1500 (CTW) [44], Total-Text (TT) [45]; (3)…”
• Scene Text Recognition using Higher Order Language Priors (2009, positive): “Each random variable x i denotes a potential character in the word, and can take a label from the label set L = { l 1 , l 2 ,..., l k }∪ e , which is the set of English characters and digits, and a null label to suppress weak detections (similar to [13]).”
• RobustScanner: Dynamically Enhancing Positional Clues for Robust Text Recognition (2020, neutral): “Most of traditional methods for scene text recognition [48,49,33,23,34] adopt the bottom-up approach in which individual character is first detected by sliding window, and then integrated for taking the dependence with its neighbors into consideration.”
• A Unified Framework for Multioriented Text Detection and Recognition (2014, methodology): “Driven by the wide range of potential applications, scene text detection and recognition have been important research topics for decades [7], [8], [35]–[41] and have recently seen a surge in research efforts in the computer vision community [9]–[14], [17], [18], [20], [21], [26], [42].”
• TextProposals: A text-specific selective search algorithm for word spotting in the wild (2016, methodology): “[15] have proposed a closely related end-toend method based on HOG features and a SVM classifier.”
• Mask TextSpotter: An End-to-End Trainable Neural Network for Spotting Text with Arbitrary Shapes (2018, neutral): “IIIT5k-Words (IIIT5k) [53] contains 3000 test images.”
• Recognizing Text with Perspective Distortion in Natural Scenes (2013, methodology): “Thus, we classify them into text MSERs and non-text MSERs using four features: the relative height, the aspect ratio, the number of holes and the number of horizontal crossings [11, 12].”