Presentation on theme: "VALSE 2014 1 Robust Scene Text Detection with Adaptive Clustering Xu-Cheng Yin ( 殷绪成 ) PhD Pattern Recognition and Information."— Presentation transcript:
VALSE 2014 1 Robust Scene Text Detection with Adaptive Clustering Xu-Cheng Yin ( 殷绪成 ) PhD http://prir.ustb.edu.cn/yin Pattern Recognition and Information Retrieval Lab Department of Computer Science and Technology University of Science and Technology Beijing 2014-09
http://prir.ustb.edu.cn/yin/ 2 Text detection in natural scenes: Background Challenges with scene text detection Complex background Variations of font and size Variations of text color Variations of illumination Variations of text orientation
http://prir.ustb.edu.cn/yin/ 3 Text detection in natural scenes: Review Previous text detection technologies Region-based (Sliding window-based) K. Kim et al., “Texture-based approach for text detection in images using SVM …”, TPAMI 2003. X. Chen, and A. Yuille, “Detecting and reading text in natural scenes”, CVPR 2004. T. Wang, D.J. Wu, A. Coates, and A. Y. Ng, “End-to-end text recognition with CNN”, ICPR 2012. VERY SLOW (each pixel, multi-scales) Connected components-based B. Epshtein et al., “Detecting text in natural scenes with stroke width transform (SWT)”, CVPR2010. C. Yao, X. Bai et al., Detecting texts of arbitrary orientations in natural images…, CVPR 2012, TIP 2014. W. Huang et al., Text localization in natural images with Stroke Feature Transform …, ICCV 2013, ECCV 2014. C. Yi and Y. Tian, Text string detection from natural scenes with boundary clustering, stroke segmentation, structure modeling, …, TIP 2011, TIP 2012, CVIU 2013. Y.-F. Pan, X. Hou and C.-L. Liu, “A hybrid approach to detect and localize texts in natural scene images”, TIP 2011 Frangibility in CC calculation
http://prir.ustb.edu.cn/yin/ 4 Text detection in natural scenes: Review Recent MSER/ER-based text detection technologies Maximally Stable Extremal Region (MSER/ER) Robust to color, size, illumination, resolution MSER/ER-based detection A specific category of CC-based methods; Use MSERs/ERs as character candidates (have become the focus of recent projects). L. Neumann and J. Matas, (Realtime) Text localization and recognition in real-world images, ACCV 2010, ICDAR 2011/2013, CVPR 2012, ICCV 2013. H.I. Koo and D.H. Kim, “Scene text detection via connected component clustering and nontext filtering”, TIP 2013. C. Shi, C. Wang, B. Xiao, et al., Scene text detection using graph model, MSER, CRF, …, Pattern Recognition Letters 2013, CVPR 2013, ICDAR 2013, TCSVT 2014, PR 2014. L. Sun, Q. Hou, et al., Robust text detection in natural scene images by Generalized Color enhanced contrasting extremal region, … ICPR 2012, ICDAR 2013, ICPR 2014. L. Kang, D. Doermann, et al., Orientation robust text line detection with HOCC…, CVPR 2014. X.-C. Yin, et al., “Robust text detection in natural scenes,” TPAMI 2014.
http://prir.ustb.edu.cn/yin/ 5 Text detection in natural scenes: Motivation Main pitfalls for MSER/ER-based text detection methods Most of the detected character candidates (MSERs/ERs) correspond to non-characters (MSER pruning) Insufficient text candidates construction with time consuming and error pruning (parameter tuning with rule-based methods) (Adaptive hierarchical clustering with metric learning) Text candidate classifier trained on an unbalanced data (Eliminating most non-text candidates with the character classifier)
http://prir.ustb.edu.cn/yin/ 6 Text detection in natural scenes: System overview
http://prir.ustb.edu.cn/yin/ 7 Text detection in natural scenes: Highlights A MSERs pruning algorithm with minimizing regularized variations is proposed to reduce most of the non-characters Character candidates are clustered into text candidates by the adaptive single-link clustering algorithm where distance weights and threshold are learned simultaneously using a self-training metric learning algorithm The posterior probabilities of text candidates corresponding to non-text are measured using the character classifier and text candidates with high probabilities for non-text are removed efficiently
http://prir.ustb.edu.cn/yin/ 8 Text detection in natural scenes: Key technologies Character candidates extraction with MSER pruning Text candidates construction with adaptive hierarchical clustering and distance metric learning Text candidates elimination with the character classifier
http://prir.ustb.edu.cn/yin/ 9 Character Candidates Extraction
http://prir.ustb.edu.cn/yin/ 10 Character Candidates Extraction (Variation regularization)
http://prir.ustb.edu.cn/yin/ 11 Text Candidates Construction Clustering-based text candidates grouping from character candidates (MSERs) Clustering: single-link clustering (elongated clusters) Similarity: weighted distance Threshold: threshold for deciding the number of clusters
http://prir.ustb.edu.cn/yin/ 12 Adaptive single-link clustering with distance metric learning Feature space (similarity)
http://prir.ustb.edu.cn/yin/ 13 Adaptive single-link clustering with distance metric learning Weighted distance Clusters How to select weights and threshold? Rule-based: time consuming and error-prone Clustering-based: a separate two-stage learning style (first weights, then threshold) Adaptive (single-link) clustering where distance weights and threshold are learned simultaneously using a self-training metric learning algorithm.
http://prir.ustb.edu.cn/yin/ 14 Adaptive single-link clustering with distance metric learning (1) Sample selection Focus on the hardest part (closest and farthest data)
http://prir.ustb.edu.cn/yin/ 16 Adaptive single-link clustering with distance metric learning (3) Model determination With the logistic regression loss, a discriminative model is designed by Distance metric learning:
http://prir.ustb.edu.cn/yin/ 18 Text Candidates Elimination Empirical results In ICDAR 2011 competition training set, only 9% of the text candidates correspond to true text Hard to train an effective text classifier using such unbalanced dataset Text candidates elimination Most methods based on rules and heuristics Our discriminative method Use a character classifier to estimate the posterior probabilities of text candidates corresponding to non-text Remove candidates with high probability for non-text
http://prir.ustb.edu.cn/yin/ 19 Text Candidates Elimination
http://prir.ustb.edu.cn/yin/ 20 Experiments On the ICDAR 2011 Robust Reading Competition Set (Challenge 2: Reading Text in Scene Images) 1,2,3,4 1.http://robustreading.opendfki.de/wiki/SceneTexthttp://robustreading.opendfki.de/wiki/SceneText 2.Top 4 winners of ICDAR2011: Kim’s, Yi’s, TH-TextLoc System, and Neumann’s 3.Shi et al.’s (Pattern Recognition Letters, 2013(2)) 4.Neuman and Matas’s (CVPR2012)
http://prir.ustb.edu.cn/yin/ 21 Experiments Speed on ICDAR 2011 data set MethodsTime (s) per imageRemarks Our Method0.43A Linux laptop with Intel (R) Core (TM)2 Duo 2.00GHZ CPU Shi et al.’s1.5A PC with Intel (R) Core (TM)2 Duo 2.33GHZ CPU Neuman and Matas’s 1.8A “standard PC”
http://prir.ustb.edu.cn/yin/ 22 Experiments (ICDAR 2011 Samples) Notice the robustness against low contrast, complex background and font variations.
http://prir.ustb.edu.cn/yin/ 23 Experiments On a publicly multilingual (include Chinese and English) dataset 1,2,3 Scheme III: constructed on ICDAR 2011 training set Scheme IV: constructed on the multilingual training set 1.http://liama.ia.ac.cn/wiki/projects:pal:member:yfpanhttp://liama.ia.ac.cn/wiki/projects:pal:member:yfpan 2.Pan et al.’s method ( Yifeng Pan, Xinwen Hou, and Cheng-Lin Liu, IEEE TIP 20(3), 2011 ) 3.Speed of Pan et al.'s method is with a PC with Pentium D 3.4GHz CPU
http://prir.ustb.edu.cn/yin/ 29 ICDAR 2013 Robust Reading Competition Results Results for the ICDAR 2013 Robust Reading Competition (Challenge2: Text Localization in Real Scenes)
http://prir.ustb.edu.cn/yin/ 30 ICDAR 2013 Robust Reading Competition Results Results for the ICDAR 2013 Robust Reading Competition (Challenge1: Text Localization in Born-Digital Images (Web and Email))
http://prir.ustb.edu.cn/yin/ 31 Main References  Xu-Cheng Yin, Xuwang Yin, Kaizhu Huang, and Hong-Wei Hao, “Robust text detection in natural scene images,” IEEE Trans. Pattern Analysis and Machine Intelligence (TPAMI), 36(5): 970-983, 2014.  Xu-Cheng Yin, Wei-Yi Pei, Jun Zhang, and Hong-Wei Hao, “Multi- orientation scene text detection with adaptive clustering”, IEEE TPAMI, submitted (with revision), 2014.  Xu-Cheng Yin, Xuwang Yin, Kaizhu Huang, and Hong-Wei Hao, “Accurate and robust text detection: A step-in for text retrieval in natural scene images”, ACM SIGIR’13.  Xuwang Yin, Xu-Cheng Yin, et al., “Effective text localization in natural scene images with MSER, geometry-based grouping and AdaBoost”, IAPR ICPR’12.
http://prir.ustb.edu.cn/yin/ 32 Discussions and Questions Industrial R&D Multilingual text detection and recognition in natural scenes, web images, ubiquitous documents and videos Academic Research End-to-end text recognition and retrieval in natural scenes and web images with Feedforward- Feedback