YunKoo Chung

Scene text extraction in natural scene images using hierarchical feature combining and verification

We propose a method that extracts text regions in natural scene images using low-level image features and that verifies the extracted regions through a high-level text stroke feature. Then the two level features are combined hierarchically. The low-level features are color continuity, gray-level variation and color variance. The color continuity is used since most of the characters in a text region have the same color, and the gray-level variation is used since the text strokes are distinctive to the background in their gray-level values. Also, the color variance is used since the text strokes are distinctive in their colors to the background, and this value is more sensitive than the gray-level variations. As a high level feature, text stroke is examined using multi-resolution wavelet transforms on local image areas and the feature vector is input to a SVM (support vector machine) for verification. We tested the proposed method with various kinds of the natural scene images and confirmed that extraction rates are high even in complex images.

An embedded camera lens distortion correction method for mobile computing applications

This paper describes a lens geometric and photometric distortion correction method to obtain a high quality image with low cost digital cameras. Lens geometric distortion coefficient of a digital camera is estimated using a simplified camera calibration technique, which is extended to calibration-free estimation method. Based on the estimate of the lens geometric distortion coefficient, image warping adapted for DSP architecture is applied. Photometric distortion, on the other hand, is corrected by a nonlinear model fitting of a proposed photometric distortion model function. Light intensity profile distribution is obtained without precisely controlled illumination and memory requirement is considerably reduced compared with traditional lookup table (LUT) methods.

Vignetting distortion correction method for high quality digital imaging

Vignetting refers to a position dependent loss of light in the output of an optical system causing gradual fading out of an image near the periphery. In this paper, we propose a method for correcting vignetting distortion by introducing nonlinear model fitting of a proposed vignetting distortion function. The proposed method aims for embedded digital imaging applications as well as conventional image enhancement. We could obtain a light intensity profile distribution without precisely controlled illumination and considerably reduce memory requirement for vignetting correction compared with the traditional lookup table (LUT) methods. We show the effectiveness of the proposed method by presenting the experimental results obtained from commercial digital cameras.

Text region extraction and text segmentation on camera-captured document style images

In this paper, we propose a text extraction method from camera-captured document style images and propose a text segmentation method based on a color clustering method. The proposed extraction method detects text regions from the images using two low-level image features and verifies the regions through a high-level text stroke feature. The two level features are combined hierarchically. The low-level features are intensity variation and color variance. And, we use text strokes as a high-level feature using multi-resolution wavelet transforms on local image areas. The stroke feature vector is an input to a SVM (support vector machine) for verification, when needed. The proposed text segmentation method uses color clustering to the extracted text regions. We improved K-means clustering method and it selects K and initial seed values automatically. We tested the proposed methods with various document style images captured by three different cameras. We confirmed that the extraction rates are good enough to be used in real-life applications.

Real-time face verification using multiple feature combination and a support vector machine supervisor

This paper proposes a novel face verification algorithm based on multiple feature combination and a support vector machine. The main issue in face verification is to deal with the variability in appearance. It seems difficult to solve this issue using a single feature. Therefore, a combination of mutually complementary features is necessary to cope with various changes in appearance. From this point of view, we describe the feature extraction approaches based on multiple principal component analysis and edge distribution. These features are projected on a new intra-person/extra-person similarity space that consists of several simple similarity measures, and are finally evaluated by a support vector machine supervisor. From the experiments on a realistic and large database, an equal error rate of 0.029 is achieved, which is a sufficiently practical level for many real-world applications.

A calibration-free lens distortion correction method for low cost digital imaging

This paper describes a new lens distortion correction method suitable for consumer digital imaging and consequent mobile computing applications. First, a convenient way of camera calibration for lens distortion coefficient estimation is proposed with experimental validation. An image warping method for lens distortion removal is formulated by inverse spatial mapping; subsequently, an approximation scheme is proposed which is adapted for real-time application. Finally, the method is extended such that the calibration stage can be completely removed. Experimental results show that the proposed method effectively replaces costly calibration process and provides a distortion-free quality image.

Distortion corrections for better character recognition of camera-based document images

The usage of cellular camera phones and digital cameras is rapidly increasing, but camera imaging applications are not so much due to the lack of practical camera imaging technology. Especially the acquisition environments of camera images are very different from those of scanner images. Light illumination, viewing distance and viewing angles constantly varies when we take a picture in indoor and outdoor. These variations make it difficult to extract character areas from images through binarization and the variation of camera viewing angles makes the images distorted geometrically. In this paper, these problems are totally discussed and the resolving methods are suggested for a better image recognition. The solutions such as adaptive binarization, color conversion, correction of lens distortion and correction of geometrical distortion are discussed and the sequence of correction processes are suggested for accurate document image recognition. In experiment, we use the various types of document images captured by mobile phone cameras and digital cameras. The results of distortion correction show that our image processing methods are efficient to increase the accuracy of character recognition for camera based document image.

Appearance-based Face Recognition from robot camera images with Illumination and Distance Variations

This paper is concerned with the appearance-based face recognition from robot camera images with illumination and distance variations. The approaches used in this paper consist of eigenface, fisherface, and icaface, which are the most representative recognition techniques frequently used in conjunction with face recognition. These approaches are based on a popular unsupervised and supervised statistical technique that supports finding useful image representations, respectively. Thus we focus on the performance comparison from robot camera images with unwanted variations. The comprehensive experiments are completed for two face databases with illumination and distance variations. A comparative analysis demonstrates that ICA comes with improved classification rates when compared with other approaches such as eigenface and fisherface

An Image Warping Method for Implementation of an Embedded Lens Distortion Correction Algorithm

Most of low cost digital cameras reveal relatively high lens distortion. The purpose of this research is to compensate the degradation of image quality due to the geometrical distortion of a lens system. The proposed method consists of two stages : calculation of a lens distortion coefficient by a simplified version of Tsai´s camera calibration and subsequent image warping of the original distorted image to remove geometrical distortion based on the calculated lens distortion coefficient. In the lens distortion coefficient calculation stage, a practical method for handling scale factor ratio and image center is proposed, after which its feasibility is shown by measuring the performance of distortion correction using a quantitative image quality measure. On the other hand, in order to apply image warping via inverse spatial mapping using the result of the lens distortion coefficient calculation stage, a cubic polynomial derived from an adopted radial distortion lens model must be solved. In this paper, for the purpose of real-time operation, which is essential for embedding into an information device, an approximated solution to the cubic polynomial is proposed in the form of a solution to a quadratic equation. In the experiment, potential for real-time implementation and equivalence in performance as compared with that from cubic polynomial solution are shown.

Face image registration methods using Normalized Cross Correlation

In the many processing stage of face recognition, one of the most important parts is face detection. In the recent years, there has been much progress in face detection. Many face recognition system adopt or develop the Adaboost: Viola-Jones face detection system. But the detected face image by adaboost method has serious problems. Face regions are apt to be different at each time, and the detected face image includes the rotated face. These issues give the bad effect to the feature extraction stage of the face recognition. In order to solve these problems, many researchers normalize the scale of face through the detection of detail element of a face, especially the eye detection. Generally, the home service robot mounts the low resolution camera. As the distance between a robot and a user becomes increasing, it is very difficult to obtain the good result by those methods. In this paper, Normalized Cross Correlation is used to detect the exact face region in the low resolution face image. The experiments showed that our method using NCC give the much better face recognition rate.