Duk-Ryong Lee

Extraction of Car License Plate Regions Using Line Grouping and Edge Density Methods

In this paper we propose the double chance algorithm as an approach to car license plate extraction. The first algorithm extracts the line segments and groups them based on a set of geometrical conditions. It detects a rectangle at the plate boundary accurately. The second algorithm finds plate regions at which the vertical edges appear densest. We evaluated the double chance framework with and without the verification process. The verification process is performed using the character segmentation module. Using a real-life database collected by the speed enforcement camera, we obtained a high success rate of 99.45%, through use of the double chance approach with verification.

Hangul document image retrieval system using rank-based recognition

We constructed a full text retrieval system for the scanned Hangul document images. The system consists of three parts; preprocessing, recognition, and retrieval. The retrieval algorithm uses recognition results up to k-ranks. The algorithm is not only insensitive to the recognition errors, but also has the advantage of user controllable recall and precision. For the objective performance evaluation, we used the scanned images of Journal of Korean Information Science Society provided by KISTI. The system was shown to be practical through the evaluation of retrieval performances.

Segmentation-free word spotting using SIFT

This paper presents a method for word spotting using SIFT. The method performs well for multi-lingual document images and it works without skew correction and without any segmentation process. The algorithm for detecting words in test images is described, including the processes of detecting keypoints, matching them, and clustering matching pairs. Experiments on English and Korean document image collections showed that the proposed method is promising.

Printed Hangul Recognition with Adaptive Hierarchical Structures Depending on 6-Types

Due to a large number of classes in Hangul character recognition, it is usual to use the six-type preclassification stage. After the preclassification, the first consonent, vowel, and last consonent can be classified separately. Though each of three components has a few of classes, classification errors occurs often due to shape similarity such as `ㅔ` and `ㅖ`. So this paper proposes a hierarchical recognition method which adopts multi-stage tree structures for each of 6-types. In addition, to reduce the interference among three components, the method uses the recognition results of first consonents and vowel as features of vowel classifier. The recognition accuracy for the test set of PHD08 database was 98.96%.

Construction of Printed Hangul Character Database PHD08

The application of OCR moves from traditional formatted documents to the web document and natural scene images. It is usual that the new applications use not only standard fonts of Myungjo and Godic but also various fonts. The conventional databases which have mainly been constructed with standard fonts have limitations in applying to the new applications. In this paper, we generate 243 image samples for each of 2350 Hangul character classes which differs in font size, quality, and resolution. Additionally each sample was varied according to binarization threshold and rotational transformation. Through this process 2187 samples were generated for each character class. Totally 5,139,450 samples constitutes the printed Hangul character database called the PHD08. In addition, we present the characteristics and recognition performance by an commercial OCR software.