Yeul-Min Baek

Image Analysis System for Measuring the Thickness of Train Brakes

In this paper we propose the system that measures the thickness of train brakes automatically. The system consists of two modules: image acquisition and analysis module that measures the thickness of brakes with digitized images. In order to measure the thickness of brake shoes, the candidate curves of brake were searched and fitted with 2d curve fitting method. To detect the disk lining brake, the edge of brake was searched first within the region of interest. Experimental results indicated that the proposed method accurately measures the thickness of brake shoe and disk lining brake with an accuracy of 0.65mm and 1.16mm, respectively at the distance of 1.0m from the camera.

Reducing overfitting of AdaBoost by clustering-based pruning of hard examples

In order to solve the problem of overfitting in AdaBoost, we propose a novel AdaBoost algorithm using K-means clustering. AdaBoost is known as an effective method for improving the performance of base classifiers both theoretically and empirically. However, previous studies have shown that AdaBoost is prone to overfitting in overlapped classes. In order to overcome the overfitting problem of AdaBoost, the proposed method uses K-means clustering to remove hard-to-learn samples that exist on overlapped region. Since the proposed method does not consider hard-to-learn samples, it suffers less from the overfitting problem compared to conventional AdaBoost. Both synthetic and real world data were tested to confirm the validity of the proposed method.

Integrated Noise Modeling for Image Sensor Using Bayer Domain Images

Most of image processing algorithms assume that an image has an additive white Gaussian noise (AWGN). However, since the real noise is not AWGN, such algorithms are not effective with real images acquired by image sensors for digital camera. In this paper, we present an integrated noise model for image sensors that can handle shot noise, dark-current noise and fixed-pattern noise together. In addition, unlike most noise modeling methods, parameters for the model do not need to be re-configured depending on input images once it is made. Thus the proposed noise model is best suitable for various imaging devices. We introduce two applications of our noise model: edge detection and noise reduction in image sensors. The experimental results show how effective our noise model is for both applications.

Color image enhancement using the laplacian pyramid

we present a color image enhancement method. The proposed method enhances the brightness and contrast of an input image using the low pass and band pass images in Laplacian pyramid, respectively. For color images, our method enhances the color tone by increasing the saturation adpatively according to the intensity of an input image. The major parameters required in our method are automatically set by the human preference data, therefore, the proposed method runs fully automatically without user interaction. Moreover, due to the simplicity and efficiency of the proposed method, a real time implementation and the enhanced results of the image quality was validated through the experiments on various images and video sequences.

Descreening Using HOG-based Adaptive Smoothing Filter

In this paper, a novel descreening method using a histogram of oriented gradients (HOG) based on an adaptive smoothing filter is proposed. Conventional descreening methods, which are used for recovering continuous-tone from scanned halftone images, are based on low-pass filtering approach. The low-pass filter has been applied to spatial domain, frequency domain, or both to remove the high frequency halftone noise. However, clipping high frequency components tends to blur images because image details also show high frequency characteristics. Moreover, due to the large variance in halftone noise, it is difficult to preserve edges by conventional adaptive smoothing methods. Therefore, to improve the descreening effect, more care has to be taken in distinguishing between edges and halftone noise. The proposed method employs HOG to distinguish edges. The amount of smoothing to be performed on the halftone image is then calculated according to the magnitude of the HOG in the edge and edge normal orientation. The proposed method was tested on various scanned halftone materials, and the results show that it removes halftone noise as effectively as the Moiré pattern while still preserving image details.

An Improved AdaBoost Algorithm by Clustering Samples

We present an improved AdaBoost algorithm to avoid overfitting phenomenon. AdaBoost is widely known as one of the best solutions for object detection. However, AdaBoost tends to be overfitting when a training dataset has noisy samples. To avoid the overfitting phenomenon of AdaBoost, the proposed method divides positive samples into K clusters using k-means algorithm, and then uses only one cluster to minimize the training error at each iteration of weak learning. Through this, excessive partitions of samples are prevented. Also, noisy samples are excluded for the training of weak learners so that the overfitting phenomenon is effectively reduced. In our experiment, the proposed method shows better classification and generalization ability than conventional boosting algorithms with various real world datasets.

Halftone Noise Removal in Scanned Images using HOG based Adaptive Smoothing Filter

In this paper, a novel descreening method using HOG(histogram of gradient)-based adaptive smoothing filter is proposed. Conventional edge-oriented smoothing methods does not provide enough smoothing to the halftone image due to the edge-like characteristic of the halftone noise. Moreover, clustered-dot halftoning method, which is commonly used in printing tends to create Moire pattern because of the intereference in color channels. Therefore, the proposed method uses HOG to distinguish edges and the amount of smoothing to be performed on the halftone image is then calculated according to the magnitude of the HOG in the edge and edge normal orientation. The proposed method was tested on various scanned halftone materials, and the results show that it effectively removes halftone noises as well as Moire pattern while preserving image details.

Measurement of the position of the overhead electric-railway line using the stereo images

In this paper, we propose a method that measures the height and stagger of an overhead electric-railway line using the stereo images. Two 1624 × 1236 pixel area scanner CCD cameras are used. To quickly and accurately extract, from a photographed image, the area of the overhead line on which the line laser is shone, we consider the established fact that the overhead line is the lowest among the electric wires. And to precisely measure the height and stagger in low resolution, sub-pixel and line fitting methods are used. Also, because of the different pixel resolution of the camera according to the overhead line position, we compensate the measurement result through camera calibration. We aimed for a measurement accuracy of 1mm error and indeed our experimental results show that the proposed method achieves that.