Young Shik Moon

A robust wavelet-based digital watermarking using level-adaptive thresholding

In this paper, a robust digital watermarking algorithm using wavelet transform is proposed. In order to equally embed the watermark to the whole image, the coefficients of all subbands including LL subband are utilized. To select perceptually significant coefficients for each subband, a level-adaptive thresholding scheme is used. The watermark is embedded to the selected coefficients, using different scale factors depending on the level of decomposition. In the detection process, the vector projection method is used. The performance of the proposed algorithm compared with other transform-domain watermarking methods has been analyzed, using various rest images. The experimental results have shown that the proposed watermark is invisible to human eyes and very robust to various attacks, such as image compression, image filtering, geometric transformations and noises.

Visual inspection system for the classification of solder joints

Abstract In this paper, efficient techniques for solder joint inspection have been described. Using three layers of ring-shaped LEDs with different illumination angles, three frames of images are sequentially obtained. From these images the regions of interest (soldered regions) are segmented, and their characteristic features including the average gray level and the percentage of highlights–referred to as 2D features — are extracted. Based on the backpropagation algorithm of neural networks, each solder joint is classified into one of the pre-defined types. If the output value is not in the confidence interval, the distribution of tilt angles — referred to as 3D features — is calculated, and the solder joint is classified based on the Bayes classifier. The second classifier requires more computation while providing more information and better performance. The choice of a combination of neural network and Bayes classifier is based on the performance evaluation of various classifiers. The proposed inspection system has been implemented and tested with various types of solder joints in SMDs. The experimental results have verified the validity of this scheme in terms of speed and recognition rate.

Design and implementation of the visual programming environment for the distributed image processing

A visual programming environment is proposed for image processing and computer vision, which is based on a dataflow model. A reusable GUI environment is designed by separating user interface and algorithms. A scheduling algorithm is also developed for the synchronous dataflow diagram. To exploit the parallelism of dataflow diagrams, a remote procedure call (RPC) to each processing node is executed. Some practical examples using our visual environment are presented and the performance improvement by RPC is quantified as the number of servers on the network increases.

An Effective Method for Eye Detection Based on Texture Information

In this paper, we present a novel method for eye detection based on texture information, referred to as the eye filter. The proposed method is composed of the following four steps: automatic extraction of facial regions using AdaBoost, illumination normalization using local information, estimation of eye candidates by the eye filter based on texture information, and detection of eye locations using geometrical information of face. Experimental results show that the eye detection rate of the proposed method is 99.2% for 3137 face images in the CAS-PEAL database and JAFFE database.

Adaptive video watermarking utilizing video characteristics in 3D-DCT domain

In this paper, we propose an adaptive blind video watermarking method using video characteristics based on human visual system (HVS) in three-dimensional discrete cosine transform (3D-DCT) domain. In order to optimize the weight factors for watermarking, we classify the patterns of 3D-DCT cubes and the types of video segments by using the texture and motion information of 3D-DCT cubes. Then we embed an optimal watermark into the mid-range coefficients of 3D-DCT cubes by using the trained optimal weight factors. Experimental results show that the proposed method achieves better performance in terms of invisibility and robustness than the previous method under the various possible attacks such as MPEG compression, frame dropping, frame insertion and frame swapping to experimental videos.

Machine Vision System for Automatic Inspection of Bridges

In the bridge inspection, the information of cracks is important to maintain a bridge. Therefore, automatic crack detection is highly desirable for efficiency and objectivity of crack assessment. However, in general, it is very difficult to detect cracks automatically from noisy concrete surfaces. In this paper, we propose a machine vision system for automatic inspection of bridges. The proposed machine vision system can detect cracks in real time, and it has some utility functions for supervised manipulation. The performance of the proposed system has been evaluated with 100 noisy images. In terms of accuracy in detecting cracks, experimental results show that the proposed method is superior to the conventional methods for detecting cracks.

Automatic Extraction of Salient Objects using Feature Maps

In this paper, we propose a technique for extracting salient objects in images using feature maps, regardless of the complexity of images and the position of objects. In order to extract salient objects, the proposed method uses feature maps with edge and color information. We also propose a reference map created by integrating feature maps, and a combination map representing the boundaries of meaningful objects that is created by integrating the reference map and feature maps. Candidate object regions including boundaries of objects from the combination map are extracted by convex hull algorithm. Finally, by applying a segmentation algorithm on the area of candidate regions, object regions and background regions are separated, and real object regions are extracted from the candidate object regions. Experimental results show that the proposed method extracts the salient objects efficiently, with 84.3% precision rate and 81.3% recall rate.

An efficient method of estimating edge locations with subpixel accuracy in noisy images

An efficient algorithm for estimating real edge locations to subpixel values in digital imagery is described. To estimate the real boundary between two adjacent pixels, we first define a 1D edge operator based on the moment invariant. To extend it to 2D data, the edge orientation of each pixel is estimated using the LSE (least squares error) method which fits a set of pixels around the edge boundary to a line/circle equation. Then, using the pixels along the line perpendicular to the estimated edge orientation, the real boundary is estimated with subpixel accuracy. Experimental results tested on real images have shown that the proposed method is robust to local noise, while maintaining a low measurement error.

A New Snake for Hand Tracking Using Textural Information

In this paper, a new snake algorithm considering textures around object boundaries is proposed for tracking deformable objects in image sequences with complex background. In order to compare the texture of object regions, a method of region classification is proposed. The proposed snake modifies the gradient value of the given image according to the textural coefficient which is driven by comparing object regions. As a result, the proposed snake can be attracted to the correct edge, although the desired edges are weak or there are strong edges in background. Experimental results have shown that the proposed snake algorithm applied to the tracking of fingers and hands produces good performance in terms of the robustness and accuracy tracking.

Face Extraction Method Using Edge Orientation and Face Geometric Features

In this paper, we propose a face extraction method based on edge and color information. In proposed method, first, we detect the candidate face region using edge orientation information. Second, we select the final face region using template which is constructed from average face. And then, we find the skin color range in the face region, and using it, process the thresholding procedure for the initial face shape. Finally, we improve face shape using the radius histogram of face shape and face model information, and then, we extract the face images. The experimental results show that the extraction performance of the proposed method is 90% of correctness.