Signboard Images Research Articles

방향성분 특징과 Fisher Measure를 이용한 간판영상 한글인식

본 논문은 모바일 폰 기반의 간판 영상내 한글 문자인식에 관한 연구로써 인식 대상은 간판영상에서 추출된 상호명으로 하였고 인식대상 문자 수는 상호명 빈도수 기반 808자로 한정하였다. 인식과정은 특징 추출, 대분류, 상세 분류로 구성되고 특징 추출과정에서는 문자영상의 크기, 잡음 및 왜곡에 강건한 비선형 방향성분 특징을 이용하였고 대분류 과정에서는 추출된 특징과 인식 대상문자에 대하여 최소거리 분류를 수행하고 10순위까지의 후보 문자를 추출하였다. 상세 분류 과정에서는 Fisher discriminant measure 이용하여 대분류에서 발생 할 수 있는 오인 식 결과를 보완하였다. 실험결과 1순위 인식률은 80.45%이고 5순위까지의 누적 인식률은 93.51%를 보였다. In this paper, we propose a Korean character recognition method from outboard signboard images. We have chosen 808 classes of Korean characters by an analysis of frequencies of appearance in a dictionary of signboard names. The proposed method mainly consists of three steps: feature extraction, rough classification, and coarse classification. The first step is to extract a nonlinear directional segments feature, which is immune to the distortion of character shapes. The second step computes an ordered set of 10 recognition candidates using a minimum distance classifier. The last step reorders the recognition candidates using a Fisher discriminant measure. As experimental results, the recognition accuracy is 80.45% for the first choice, and 93.51% for the top five choices.

Image Processing/Conding and Applications: Thinning/Thickening and Difference Sub-template Matching-For Robust Recognition of Strongly Deteriorated Signboard Images-

In this study, we developed a novel template matching method by establishing two new types of subtemplates for robust recognition of strongly deteriorated signboard images. This method has three main stages: (i) removing occlusion/reflection by using the main peak finding algorithm of uniform color regions; (ii) establishing thinning/thickening and difference sub-templates by using morphological operations; and (iii) adopting template matching criterion based on modified standard deviation. The high performance of our proposed method is shown in the results of an experiment carried out on 200 outdoor images taken under various kinds of bad conditions with occlusion, refection and so on.

Digit recognition in a natural scene with skew and slant normalization

This paper proposes a method to recognize digits in a natural scene, such as telephone numbers on a signboard. Candidate regions of digits are extracted from an image through contrast enhancement, edge extraction, and labeling. Since the target text patterns are in a 3D space, unlike traditional character recognition problems, we have to deal with the image transformation effect due to the orientation in the 3D space and projection. We have to cancel the effect as much as possible before digit recognition. In our method, the image transformation effect is modeled as skew and slant. In the proposed method, simplified Hough transform is used for the skew normalization. After the skew normalization, the remaining effect of image transformation is corrected by circumscribing digit patterns with tilted rectangles and affine transformation. In experiments, we tested a total of 1,332 images of signboards with 11,939 digits. We obtained a digit extraction rate of 99.2% and a correct digit recognition rate of 98.8%.

A Contents Service System Using a Signboard Image Captured with Mobile-Phone Cameras as a Key

A Contents Service System Using a Signboard Image Captured with Mobile-Phone Cameras as a Key