Camera-Based Document Analysis and Recognition: 5th by Masakazu Iwamura, Faisal Shafait

By Masakazu Iwamura, Faisal Shafait

This ebook constitutes the completely refereed post-workshop complaints of the fifth overseas Workshop on Camera-Based record research and popularity, CBDAR 2013, held in Washington, DC, united states, in August 2013. The 14 revised complete papers awarded have been rigorously chosen in the course of rounds of reviewing and development from various unique submissions. meant to offer a picture of the state of the art examine within the box of digicam dependent record research and popularity, the papers are equipped in topical sections on textual content detection and popularity in scene photographs and camera-based systems.

Show description

Read or Download Camera-Based Document Analysis and Recognition: 5th International Workshop, CBDAR 2013, Washington, DC, USA, August 23, 2013, Revised Selected Papers PDF

Best human-computer interaction books

Computer Vision for Human-Machine Interaction

Fresh advances within the box of laptop imaginative and prescient are resulting in novel and radical alterations within the means we have interaction with desktops. it's going to quickly be attainable to permit a working laptop or computer associated with a video digital camera to observe the presence of clients, tune faces, hands and fingers in actual time, and examine expressions and gestures.

Plan, Activity, and Intent Recognition: Theory and Practice

Plan popularity, task acceptance, and reason reputation jointly mix and unify thoughts from consumer modeling, desktop imaginative and prescient, clever consumer interfaces, human/computer interplay, self sufficient and multi-agent structures, normal language realizing, and laptop studying. Plan, task, and rationale reputation explains the the most important position of those ideas in a large choice of functions together with: .

Online Social Media Analysis and Visualization

This edited quantity addresses the sizeable demanding situations of adapting on-line Social Media (OSM) to constructing study equipment and purposes. the themes disguise producing reasonable social community topologies, expertise of person actions, subject and development new release, estimation of person attributes from their social content material, habit detection, mining social content material for universal traits, deciding on and rating social content material assets, development friend-comprehension instruments, etc.

Additional info for Camera-Based Document Analysis and Recognition: 5th International Workshop, CBDAR 2013, Washington, DC, USA, August 23, 2013, Revised Selected Papers

Example text

Performance (%) comparison of text detection approaches on ICDAR2011 robust reading competition dataset. 63 Fig. 6. Performance under different component classifier numbers. 1 %. Figure 8 shows some detection examples, where most of the text objects are correctly detected with few false positives. The text objects are in complex background and can have low resolution or low contrast. This shows that the proposed approach can correctly capture text patterns of large variations simultaneously with an integrated discrimination.

6. In Fig. 6(a), ‘1’ represents foreground pixels and ‘0’ represents background pixels. Each object pixel of this binary image is assigned a value which is its distance from the nearest background pixel as shown in Fig. 6(b). At each non-zero pixel of D∗ , we consider a 3 × 3 window to obtain the local maximum at that pixel. If this pixel value equals the local maximum, we store the pixel value in a list < T > for further processing. In fact, such a pixel value (a local maximum value) is an estimate of half of the local stroke thickness.

The MSER algorithm adaptively detects stable color regions and provides a good solution to localize the components. In [8,10], MSERs from H, S, I and gradient channels are integrated to detect components. An exhaustive search is then applied to group components into regions and a text level classifier is used for classification of these regions. In [11], Koo et al present a text detection approach based on MSERs and two classifiers. The first classifier is trained on AdaBoost that determines the adjacency relationship and cluster components by using pairwise relations.

Download PDF sample

Rated 4.22 of 5 – based on 50 votes