Gueesang Lee

Efficient skew estimation and correction algorithm for document images

Abstract In this paper, we propose a fast skew estimation and correction algorithm for English and Korean documents based on a BAG (Block Adjacency Graph) representation. BAG is one of the most efficient data structures for extracting various information concerning connected components; the image rotation for skew correction is performed rapidly using the block information in the BAG. The proposed skew estimation algorithm uses a coarse/refine strategy based on the Hough transformation of connected components in the image. The skew correction algorithm then generates a non-skew image by rotating the blocks, rather than the individual pixels. An experiment using 2016 images from various English and Korean documents demonstrates how the proposed method is superior to conventional ones.

An efficient recovery mechanism for MPLS-based protection LSP

Network reliability and survivability have been very important issues to provide for application services that may require a real-time service or high priority QoS (quality of service) in the Internet. IETF has proposed largely two recovery mechanisms for MPLS-based protection label switching path (LSP), which are protection switching and rerouting models. However, from the viewpoint of TE (traffic engineering), IETFs recovery mechanisms have not considered the optimal backup path for the recovery of an LSP in the occurrence of a network failure. This paper suggests an efficient pre-qualified recovery mechanism, which optimizes the network performance by considering link usage. Since an existing recovery mechanism, pre-qualified rerouting, selects a backup path only once at the LSP setup time, it may not reflect the exact status of network resources at the time of a fault. In contrast, our approach exchanges network status information among LSRs so that the backup path selection engine can use up-to-date information and decide an optimal backup path for a possible failure. The performance of the proposed recovery method has been demonstrated by simulation using MNS (MPLS network simulator). The new proposed recovery mechanism can always maintain an optimized network state regardless of the fault occurrence.

Fall Detection Based on Movement and Smart Phone Technology

Nowadays, recognizing human activities is an important subject; it is exploited widely and applied to many fields in real-life, especially health care or context aware application. Research achievements are mainly focused on activities of daily living which are useful for suggesting advises to health care applications. Falling event is one of the biggest risks to the health and well being of the elderly especially in independent living because falling accidents may be caused from heart attack. Recognizing this activity still remains in difficult research area. Many systems which equip wearable sensors have been proposed but they are not useful if users forget to wear the clothes or lack ability to adapt themselves to mobile systems without specific wearable sensors. In this paper, we develop novel method based on analyzing the change of acceleration, orientation when the fall occurs. In this study, we recruit five volunteers in our experiment including various fall categories. The results are effective for recognizing fall activity. Our system is implemented on Google Android smart phone which already plugged accelerometer and orientation sensors. The popular phone is used to get data from accelerometer and results show the feasibility of our method and contribute significantly to fall detection in Health care.

Automatic detection and recognition of Korean text in outdoor signboard images

In this paper, an automatic translation system for Korean signboard images is described. The system includes detection and extraction of text for the recognition and translation of shop names into English. It deals with impediments caused by different font styles and font sizes, as well as illumination changes and noise effects. Firstly, the text region is extracted by an edge-histogram, and the text is binarized by clustering. Secondly, the extracted text is divided into individual characters, which are recognized by using a minimum distance classifier. A shape-based statistical feature is adopted, which is adequate for Korean character recognition, and candidates of the recognition results are generated for each character. The final translation step incorporates the database of shop names, to obtain the most probable result from the list of candidates. The system has been implemented in a mobile phone and is demonstrated to show acceptable performance.

Word segmentation in handwritten Korean text lines based on gap clustering techniques

We propose a word segmentation method for handwritten Korean text lines. It uses gap information to separate a text line into word units, where the gap is defined as a white-run obtained after a vertical projection of the line image. Each gap is classified into a between-word gap or a within-word gap using a clustering technique. We take up three gap metrics - the bounding box (BB), run-length/Euclidean (RLE) and convex hull (CH) distances - which are known to have superior performance in Roman-style word segmentation, and three clustering techniques - the average linkage method, the modified MAX method and sequential clustering. An experiment with 498 text-line images extracted from live mail pieces has shown that the best performance is obtained by the sequential clustering technique using all three gap metrics.

Small Object Segmentation Based on Visual Saliency in Natural Images

Object segmentation is a challenging task in image processing and computer vision. In this paper, we present a visual attention based segmentation method to segment small sized interesting objects in natural images. Different from the traditional methods, we first search the region of interest by using our novel saliency-based method, which is mainly based on band-pass filtering, to obtain the appropriate frequency. Secondly, we applied the Gaussian Mixture Model (GMM) to locate the object region. By incorporating the visual attention analysis into object segmentation, our proposed approach is able to narrow the search region for object segmentation, so that the accuracy is increased and the computational complexity is reduced. The experimental results indicate that our proposed approach is efficient for object segmentation in natural images, especially for small objects. Our proposed method significantly outperforms traditional GMM based segmentation. Keywords—Gaussian Mixture Model (GMM), Visual Saliency, Segmentation, Object Detection.

Implicit authentication based on ear shape biometrics using smartphone camera during a call

From secret knowledge like password up to physical traits as biometrics, current smartphone authentication systems are deemed inconvenience and difficult for users. Burdens on remembering password as well as privacy issues on stolen or forged biometrics have raised a futuristic idea of authentication systems. New system is hoped being transparent and with very minimum user involvement denoted as implicit authentication system. One of the ways to implicitly authenticate users is by authenticating them via image or video captured using smartphone camera during a call. During call interaction, we implicitly take ear image using front smartphone camera to recognize and authenticate users without them realizing. In this paper, we present a novel approach to ear recognition which considers both shape and texture information to represent ear image. Firstly, all Local Binary Pattern (LBP) are combined after extracted and concatenated into a single histogram. Second, in order to get geometric features, we use the idea of ear location center that is easily adjusted by smartphone user. Then, we combine previous steps to represent ear image as a descriptor. The recognition is performed using a nearest neighbor classifier computed feature space with Euclidean distance as a similarity/dissimilarity measure. Our proposed approach is very easy and simple thereby its simplicity allows very fast feature extraction. We foresee that this experiment is applicable directly on smartphone.

Tensor Voting Based Text Localization in Natural Scene Images

A new and efficient text localization method by tensor voting is proposed. Tensor voting is used to extract the text line information based on the observation that the text characters are situated close together and arranged in a line or on a smooth curve. The text line information is useful to reduce the false positive rate in region-based text localization methods. The experimental results obtained for different types of natural text images show that the proposed method successfully detects the text regions with a low false-positive rate.

Robust background extraction scheme using histogram-wise for real-time tracking in urban traffic video

Detecting moving objects in video sequence with a lot of moving vehicles and other difficult conditions is a fundamental and difficult task in many computer vision applications. A common approach is based on background extraction, which identifies moving objects from the input video frames that differs significantly from the background model. There are many challenges exist in extracting a good background. Our proposed method is able to deal with illumination change based on homomorphic filer. Using histogram- wise makes the background model be actively reacted to the changes in background such as starting and stopping of vehicles. Moreover, the identification of stationary objects such as swinging leaves, rain, snow, lanes, shadows, can make the background model more robust and also can reduce the processing time. The proposed method is simplicity and effectiveness by a simple technique to more sophisticated probabilistic modeling techniques. The algorithm has been successful tested in busy traffic and other difficult conditions scenes with the comparisons to some basic and complex probabilistic algorithms.

Side Information Generation Using Extra Information in Distributed Video Coding

We address the hash-based side information generation problem in Distributed Video Coding (DVC). The side information plays an important role in DVC because more accurate side information is generated with less number of parity bits needed to successfully decode the Wyner-Ziv (WZ) frame. In hash-based side information generation, the encoder also generates and transmits some hash codes about the current WZ frame to help improve the quality of side information, especially when using motion extrapolation techniques. In the paper, hashes are designed using various types of extra information and their impacts on side information are examined.