Adaptive Multi-Scale HyperNet with Bi-Direction Residual Attention Module for Scene Text Detection

Scene text detection and recognition are vital challenging tasks in computer vision, which are to detect and recognize sequences of texts in natural scenes. Recently, researchers have investigated a lot of state-of-the-art methods to improve the accuracy and efficiency of text detection and recognition. However, there has been little research on text detection and recognition in natural scenes in Vietnam. In this paper, a deep learning-based empirical investigation of Vietnamese scene text detection and recognition is presented. Firstly, four detection models including differentiable binarization network (DBN), pyramid mask text detector (PMTD), pixel aggregation network (PAN), and Fourier contour embedding network (FCEN), are employed to detect text regions from the images. Then, four text recognition models including convolutional recurrent neural network (CRNN), self-attention text recognition network (SATRN), no-recurrence sequence-to-sequence text recognizer (NRTR), and RobustScanner (RS) are also investigated to recognize the texts. Moreover, data augmentation methods are also applied to enrich data for improving the accuracy and enhancing the performance of scene text detection and recognition. To validate the effectiveness of scene text detection and recognition models, the VinText dataset is employed for evaluation. Empirical results show that PMTD and SATRN achieve the highest scores among the others for text detection and recognition, respectively. For knowledge-sharing, our implementation is publicly available at https://github.com/ThorPham/VN_scene_text_detection_recognition.

Scene texts refer to arbitrary text found in images captured by cameras in real-world settings. The tasks of text detection and recognition are critical components of computer vision, with applications spanning scene understanding, information retrieval, robotics, and autonomous driving. Despite significant advancements in deep learning methods, achieving accurate text detection and recognition in complex images remains a formidable challenge for robust real-world applications. Several factors contribute to these challenges. First, the diversity of text shapes, fonts, colors, and styles complicates detection efforts. Second, the myriad combinations of characters, often with unstable attributes, make complete detection difficult, especially when background interruptions obscure character strokes and shapes. Finally, effective coordination of multiple sub-tasks in end-to-end learning is essential for success. This research aimed to tackle these challenges by enhancing text discriminative representation. This study focused on two interconnected problems: Scene Text Recognition (STR), which involves recognizing text from scene images, and Scene Text Detection (STD), which entails simultaneously detecting and recognizing multiple texts within those images. This research focuses on implementing and evaluating the Efficient and Accurate Scene Text Detector (EAST) algorithm for text detection and recognition in natural scene images. The study aims to compare the performance of three prominent Optical Character Recognition (OCR) techniques—TesseractOCR, PaddleOCR, and EasyOCR. The EAST model was applied to a series of sample test images, and the results were visually represented with bounding boxes highlighting the detected text regions. The inference times for each image were recorded, highlighting the algorithm's efficiency, with average times of 0.446, 0.439, and 0.440 seconds for the respective test images. These results indicate that the EAST algorithm is accurate and operates in real-time, making it suitable for applications requiring immediate text recognition.

The existing deep learning based state-of-theart scene text detection methods treat scene texts a type of general objects, or segment text regions directly. The latter category achieves remarkable detection results on arbitraryorientation and large aspect ratios of scene texts based on instance segmentation algorithms. However, due to the lack of context information with consideration of scene text unique characteristics, directly applying instance segmentation to text detection task is prone to result in low accuracy, especially producing false positive detection results. To ease this problem, we propose a novel text-context-aware scene text detection CNN structure, which appropriately encodes channel and spatial attention information to construct context-aware and discriminative feature map for multi-oriented and multi-language text detection tasks. With high representation ability of textcontext-aware feature map, the proposed instance segmentation based method can not only robustly detect multi-oriented and multi-language text from natural scene images, but also produce better text detection results by greatly reducing false positives. Experiments on ICDAR2015 and ICDAR2017-MLT datasets show that the proposed method has achieved superior performances in precision, recall and F-measure than most of the existing studies.

The generation of text proposals (i.e. local candidate regions most likely containing textual components) is one critical and prerequisite step in scene text detection task. As one popular text proposal algorithm, the Maximally Stable Extremal Region (MSER), has been exploited by many successful text detection methods, while on the other hand has difficulties in handling complicated scene text involving touching characters and characters composed of multiple unconnected parts (e.g. Chinese characters and text in dot matrix fonts). In this paper, we propose a novel text proposal method for localizing text in natural images, which integrates the MSER algorithm with the multi-scale sliding window framework and efficiently extracts Windowed Maximally Stable Extremal Regions (WMSERs) as text proposals. We further present effective proposal filtering and grouping algorithms for exploiting WMSER-based proposals in text detection task. Experiments on public scene text datasets demonstrate the promising aspects of the proposed method in dealing with complicated scene text.

Text detection in natural scene images is a challenging problem in computer vision. To robust detect various texts in complex scenes, a hierarchical recursive text detection method is proposed in this paper. Usually, texts in natural scenes are not alone and arranged into lines for easy reading. To find all possible text lines in an image, candidate text lines are obtained using text edge box and conventional neural network at first. Then, to accurately find out the true text lines in the image, these candidate text lines are analyzed in a hierarchical recursive architecture. For each of them, connected components segmentation and hierarchical random field based analysis are recursively employed until the detected text line no more changes. Now the detected text lines are output as the text detection result. Experiments on ICDAR 2003 dataset, ICDAR 2013 dataset and Street View Dataset show that the hierarchical recursive architecture can improve text detection performance and the proposed method achieves the state-of-art in scene text detection.

As an important information carrier, texts play significant roles in many applications. However, text detection in unconstrained scenes is a challenging problem due to cluttered backgrounds, various appearances, uneven illumination, etc.. In this paper, an approach based on multi-channel information and local context is proposed to detect texts in natural scenes. According to character candidate detection plays a vital role in text detection system, Maximally Stable Extremal Regions(MSERs) and Graph-cut based method are integrated to obtain the character candidates by leveraging the multi-channel image information. A cascaded false positive elimination mechanism are constructed from the perspective of the character and the text line respectively. Since the local context information is very valuable for us, these information is utilized to retrieve the missing characters for boosting the text detection performance. Experimental results on two benchmark datasets, i.e., the ICDAR 2011 dataset and the ICDAR 2013 dataset, demonstrate that the proposed method have achieved the state-of-the-art performance.

Scene text detection is an important and challenging task in computer vision. For detecting arbitrarily-shaped texts, most existing methods require heavy data labeling efforts to produce polygon-level text region labels for supervised training. In order to reduce the cost in data labeling, we study mixed-supervised arbitrarily-shaped text detection by combining various weak supervision forms (e.g., image-level tags, coarse, loose and tight bounding boxes), which are far easier to annotate. Whereas the existing weakly-supervised learning methods (such as multiple instance learning) do not promote full object coverage, to approximate the performance of fully-supervised detection, we propose an Expectation-Maximization (EM) based mixed-supervised learning framework to train scene text detector using only a small amount of polygon-level annotated data combined with a large amount of weakly annotated data. The polygon-level labels are treated as latent variables and recovered from the weak labels by the EM algorithm. A new contour-based scene text detector is also proposed to facilitate the use of weak labels in our mixed-supervised learning framework. Extensive experiments on six scene text benchmarks show that (1) using only 10% strongly annotated data and 90% weakly annotated data, our method yields comparable performance to that of fully supervised methods, (2) with 100% strongly annotated data, our method achieves state-of-the-art performance on five scene text benchmarks (CTW1500, Total-Text, ICDAR-ArT, MSRA-TD500, and C-SVT), and competitive results on the ICDAR2015 Dataset. We will make our weakly annotated datasets publicly available.

Video scene text contains semantic information and thus can contribute significantly to video indexing and summarization. However, most of the previous approaches to detecting scene text from videos experience difficulties in handling texts with various character size and text alignments. In this paper, we propose a novel algorithm of scene text detection and localization in video. Based on our observation that text character strokes show intensive edge details in the fixed orientation no matter what text alignment and size are, a stroke map is first generated. In the scene text detection, we extract the texture feature of stroke map to locate text lines. The detected scene text lines are accurately located by using Harris' corners in the stroke map. Experimental results show that this approach is robust and can be effectively applied to scene text detection and localization in video.

Scene text detection is a challenging problem due to the image cluttering and high variability of text shape. Many methods have been proposed for multi-oriented and arbitrary shape text detection, in which the storage and computation costs of deep neural networks are still concerns. In this paper, we first introduce Octave Convolution into scene text detection for enlarging the receptive fields and reducing the spatial redundancy of networks. Then we combine Octave Convolution with a state-of-the-art arbitrary shape text detector PSENet, which predicts different scale of kernels for each text instance. Experimental results on several benchmarks show that the proposed method can improve both detection performance and speed in detecting multi-oriented and arbitrary shape texts. Furthermore, our method achieves state-of-the-art performances on these benchmarks.

Scene text in video contains semantic information and thus can contribute significantly to video retrieval and understanding. However, most methods detect scene text in image or single frame. Videos differ from images in temporal redundancy. In this paper we present a novel approach to detecting video scene text based on the video temporal redundancy. Video scene texts in consecutive frames have arbitrary motion due to camera or object movement. Therefore, first, we perform the motion detection in 30 consecutive frames to synthesize motion image. Second we implement video scene text detection in single frame to retrieve candidate text regions. Finally, the synthesized motion image is used to filter out candidate text regions and only keep the candidate text regions which have motion occurrence as final scene text. Experimental results show that our algorithm performs well for detecting scene text with various color, font-size and text alignment.

In this paper, we investigate the issue of class imbalance in scene text detection. Class Balanced Cross Entropy (CBCE) loss is often adopted for addressing this imbalance problem. We find that CBCE excessively restrains the backward gradients of background. Negative samples own extremely small weights which are offered by CBCE during training of text detectors. These tiny weight values lead to insufficient learning of background. As a result, the CBCE-based text detection methods only can achieve sub-optimal performance.We propose a novel loss function, Strong-Background Restrained Cross Entropy (SBRCE), to deal with the disadvantage in CBCE. Specifically, SBRCE effectively down-weights the loss assigned to the strong background which means well-classified negative samples. Our SBRCE can make training focused on all positive samples and weak background(i.e., hard-classified negative samples). Moreover, it can prevent the enormous amount of strong background from overwhelming text detectors during training. Experimental results show that the proposed SBRCE can improve the performance of the efficient and accurate scene text detector (EAST) by F-score of 3.3% on ICDAR2015 dataset and 1.12% on MSRA-TD500 dataset, without sacrificing the training and testing speed of EAST.

Visually impaired people suffer daily from their disability to read textual information. One of the most anticipated blind-assistive devices is a system equipped with a wearable camera capable of finding the textual information in natural scenes and translating it into sound through a speech synthesizer. To avoid duplicate readings, the device should be able to recognize text areas with the same content, and group them to obtain a single result. Scene text detection and tracking methods attract a lot of interest for these purposes. However, this field is still challenging and methods of scene text detection and tracking are yet to be perfected. This paper proposes a scene text tracking system capable of finding text regions and tracking them in video frames captured by a wearable camera. By combining a text detection method with a feature point tracker, we obtain a robust text tracker which produces much less false positive text images at 2.9 times faster speed compared with the conventional method.

Text extraction from scene images by character appearance and structure modeling

Buffer-text: Detecting arbitrary shaped text in natural scene image

Recently, scene text detection has become an active research field, which is an essential component of scene text reading. Especially, segmentation-based methods are commonly used, since the segmentation results can describe text of arbitrary shape. However, curve texts have a diversity of shapes, scales and orientations, which are difficult to locate, so the detector requires to adjust the local receptive fields size adaptively, which can aggregate multi-scale spatial information to accurately locate the curve text instance. Moreover, the low-level features are critical for localizing large text instances. When using Feature Pyramid Network (FPN) for multi-scale feature fusion, it will prevent the flow of accurate localization signals due to the long path from low-level to top-level. In order to solve these two problems, this paper proposes an Adaptive Convolution and Path Enhancement Pyramid Network (ACPEPNet), which can more accurately locate the text instances with arbitrary shapes. Firstly, an Adaptive Convolution Unit is introduced to improve the ability of backbone to aggregate multi-scale spatial information at the same stage. Specially, this unit is a lightweight component and without the cost of computations, based on this component we present a backbone network for text features extraction. Secondly, the original FPN structure is redesigned to build a short path from the low-level to top-level, in this way, we modify the path from one-way flow to two-way flow and add original features to the final stage of information fusion. Experiments on CTW1500, Total-Text, ICDAR 2015 and MSRA-TD500 validate the robustness of the proposed method. When there is no bells and whistles, this method achieves an F-measure of 80.8% without external training data on CTW1500.