Accurate junction detection and characterization in line-drawing images
Introduction
A junction point, by definition in the computer vision (CV) field [1], is formed by the intersection of at least two homogeneous regions. As a result, the junctions are often detected by finding the prominent points in the image at which the boundaries of the adjacent regions meet. The edges meeting at a junction point are regarded as the arms of the junction and are used to characterize junctions into different types such as L-, T-, or X-junctions. Even though much of work for junction detection has been proposed in CV [2], [3], [4], [5], [6], it is difficult to directly apply these methods for the same problem in document image analysis (DIA) for several reasons. First, most of these methods are scale-dependent, and restricted to limited experiments without comparative evaluation with other methods. Few [2], [6] discuss junction characterization, which is very important for junction features. Second, all of these methods are limited to single junction detection. Finally, the conception of a junction in DIA is different from that in CV, making CV methods unsuitable.
In DIA, the junction points are treated as the intersections of at least two line segments and the problem of junction detection is usually formalized as finding the intersections of median lines in images. This point, in fact, constitutes the main challenge of junction detection in DIA as the task of extracting the median lines is not trivial. For these reasons, dedicated methods [7], [8], [9], [10], [11] have been proposed in DIA, in which the junction detection problem has been mainly considered as a post-processing of a vectorization process. Although most of these well-known techniques for junction detection are vectorization-based systems, such methods rely on vectorization, which is known to be sensitive to setting parameters, and present difficulties when heterogeneous primitives (e.g., straight lines, arcs, curves, and circles) appear within a single document. Knowledge about the document content must be included, making the systems less adaptable to heterogeneous corpus.
In this work, we directly address the problem of junction detection by searching for optimal meeting points of median lines. At first glance, it seems that our approach would directly encounter the well-known problem of junction distortion. However, it is important to note that, apart from distorted zones (i.e., the areas where several line segments meet), the median lines are known to be very representative for the rest of line segments. This point suggests that if we can successfully remove the distorted zones, the remaining disjointed strokes would be not subjected to the problem of junction distortion. We therefore present a new algorithm to precisely detect and conceptually remove the distorted zones. The remaining line segments are then locally characterized to form structural representations of the crossing zones. Finally, an optimization algorithm is proposed to reconstruct the junctions.
We review the related work for junction detection in Section 2. The details of the proposed approach are presented in Section 3. Junction characterization and matching are presented in Section 4. A complexity evaluation of the proposed system is given in Section 5. Experimental results are investigated in Section 6. An application of symbol spotting is provided in Section 7. Key remarks and future works are given in Section 8.
Section snippets
Junction detection techniques in CV
The methods for junction detection in CV are often classified into two categories: edge grouping and template matching [1]. For the former approaches of edge grouping, Bergevin and Bubel [2] first detect edge points by applying several local oriented-based filters. The detected edge points are then used in conjunction with two criteria (i.e., spatial dispersion and occupancy rate) to generate hypotheses about junction branches from which junction points will be constructed. Deschênes and Ziou
The proposed approach
We directly formulate the problem of junction detection as searching for optimal meeting points of line-like primitives from input images. However, as it is impossible to obtain ideal line primitives (i.e., 1-pixel-thick lines) from a digitalization process, the intersection areas of the line primitives can not converge or contract to one pixel as expected. Therefore, to achieve exact junction localization, the line primitives must be represented in suitable forms that facilitate the step of
Junction characterization
One of the main advantages of our junction reconstruction process is that the detected junctions could be automatically characterized and classified into different types, such as T-, L-, and X-junctions. More generally, we wish to characterize any complicated junctions in the same manner based on the arms forming the junction. In our case, as each junction point is constructed from the local line segments of one group, we can consider these line segments as the arms of the junction point.
Complexity evaluation
In this section, we provide a detailed analysis of the complexity of the proposed method given an image I of the size M×N. In the pre-processing stage, before applying the (3,4)-distance transform skeletonization algorithm, several basic pre-processing steps, such as hole filling, small contour removing, and image dilation, are performed, as discussed in the original work of Di Baja [20]. Such steps can be processed in parallel using one scan over the image. The skeletonization step is then
Evaluation metric and protocol
We use repeatability criterion to evaluate the performance of our junction detector because this criterion is standard for the performance characterization of local keypoint detectors in CV [28]. This criterion works as follows. Given a reference image Iref and a test image Itest taken under different transformations (e.g., noise, rotation, scaling) from Iref, the repeatability criterion signifies that the local features detected in Iref should be repeated in Itest with some small error in
Application to symbol spotting
We present here additional experiments at the application level to prove that our junction detector is robust and discriminant enough to be used in a retrieval, spotting, or recognition systems. We have used our approach in a baseline symbol spotting system. A symbol spotting system is often composed of the following modules: the decomposition of document images into primitives; the characterization of the primitives; primitive matching, grouping and localization; and verification. In our case,
Conclusions and future works
In this paper, a new approach for junction detection and characterization in line-drawings has been presented. The main contribution of this work is three-fold. First, a new algorithm for the determination of the region of support is presented using the linear least squares technique. The crossing-points, in combination with the dominant points detected from median lines, are treated as candidate junctions. Next, using these candidate junctions, an efficient algorithm is proposed to detect and
Acknowledgments
The authors would like to thank the anonymous reviewers for their valuable comments and suggestions on this manuscript. This work has been supported by the Viet Nam International Education Development project (VIED).
Conflict of interest
None declared.
The Anh Pham received the Masters (Research) with specialization in Computer Science from Vietnam National University, Hanoi in 2006 (Vietnam). He has then worked as a lecturer at Hong Duc University (Vietnam) since 2007. In 2010, he was awarded a scholarship from Vietnam International Education Development (VIED) for pursuing his Ph.D. research in France.
References (30)
- et al.
Detection and characterization of junctions in a 2d image
Comput Vision and Image Understanding
(2004) - et al.
Detection of line junctions and line terminations using curvilinear features
Pattern Recognition Letters
(2000) - et al.
A new algorithm for line image vectorization
Pattern Recognition
(1998) - et al.
A chinese-character-stroke-extraction algorithm based on contour information
Pattern Recognition
(1998) - et al.
A run-length-coding-based approach to stroke extraction of chinese characters
Pattern Recognition
(2000) Well-shaped, stable, and reversible skeletons from the (3,4)-distance transform
Journal of Visual Communication and Image Representation
(1994)- et al.
Performance evaluation of corner detectors using consistency and accuracy measures
Computer Vision and Image Understanding
(2006) - et al.
Dominant point detectiona new proposal
Image and Vision Computing
(2005) Generalizing the hough transform to detect arbitrary patterns
Communications of the ACM
(1981)- et al.
Junctionsdetection, classification, and reconstruction
IEEE Transactions on Pattern Analysis and Machine Intelligence
(1998)
Sparse pixel vectorizationan algorithm and its performance evaluation
IEEE Transactions on Pattern Analysis and Machine Intelligence
Robust and accurate vectorization of line drawings
IEEE Transactions on Pattern Analysis and Machine Intelligence
Cited by (25)
Junction detection in handwritten documents and its application to writer identification
2015, Pattern RecognitionCitation Excerpt :Generally, the coherence of the detection through some transformations, such as scaling and rotation, is investigated under some criterion. The repeatability criterion is widely used for performance characterization both in junction detection [8,13] and local keypoint detectors in computer vision [48]. In this paper, we follow the evaluation method proposed in [8,48].
Sketch-based interaction and modeling: Where do we stand?
2019, Artificial Intelligence for Engineering Design, Analysis and Manufacturing: AIEDAMGolden ratio in venation patterns of dragonfly wings
2023, Scientific ReportsImage Feature Information Extraction for Interest Point Detection: A Comprehensive Review
2023, IEEE Transactions on Pattern Analysis and Machine IntelligenceSimplification and vectorization of sketched drawings
2022, Interactive Sketch-Based Interfaces and Modelling for DesignGolden ratio in venation patterns of dragonfly wings
2022, Research Square
The Anh Pham received the Masters (Research) with specialization in Computer Science from Vietnam National University, Hanoi in 2006 (Vietnam). He has then worked as a lecturer at Hong Duc University (Vietnam) since 2007. In 2010, he was awarded a scholarship from Vietnam International Education Development (VIED) for pursuing his Ph.D. research in France.
Mathieu Delalandre obtained two M.Sc. in Computer Science and Electrical Engineering from the Rouen University (France) in 2001. During the years 2001–2005, he has done his Ph.D. Thesis in the same University at the LITIS Laboratory. Then, starting from 2006 until 2009 he has done different Research Fellow positions in laboratories and institutes Europe-wide: SCSIT (Nottingham, UK), L3i (La Rochelle, France) and CVC (Barcelona, Spain). Since September 2009 he is an Assistant Professor at the LI laboratory (Tours, France) in the RFAI group.
Sabine Barrat received her Ph.D. in Computer Science (2009) from Nancy 2 University (France). The subject of her Ph.D. Thesis was about modeling, retrieval, classification and automatic annotation of images, by combining visual and semantic information with probabilistic graphical models. After her Ph.D., she was awarded a JSPS Postdoctoral Fellowship and spent 6 months in the Intelligent Media Processing group of Osaka Prefecture University, Japan. Since 2010, she is an Assistant Professor at the LI laboratory (Tours, France) in the RFAI group.
Jean-Yves Ramel received his Ph.D. in Computer Science (1996) from the RFV/LIRIS Laboratory in Lyon (France). From 1998 to 2002, he was working in the field of Man–Machine Interaction at INSA Lyon. Since 2002, he is working in the field of Pattern Recognition and Image Analysis at the Computer Sciences Laboratory (LI) of Tours (RFAI team) at Polytech'Tours (France). Since September 2007, he is Professor at the LI laboratory in the RFAI group.