Sung-Il Chien

Automatic car license plate extraction using modified generalized symmetry transform and image warping

This paper proposes a method which evaluates symmetry of plate corners and extracts car license plates captured from the arbitrary directions. The generalized symmetry transform (GST) produces continuous features of symmetry between two points by combining locality constraint and reflectional symmetry. Also, we propose scan line based GST to improve time complexity of the GST significantly due to selective attention to particular distortions in case of a large searching window. The corners of a plate are detected from our modified scan line based GST with selective directions using the fact that a corner is formed by two neighboring straight lines acquired in a symmetry map of GST. However, a rotated or perspectively distorted car license plate image is very difficult for segmentation of interior characters. Image normalization by image warping is adopted to make such segmentation of license plate and later identification much easier. We also adopt the verifier which evaluates a candidate license plate to enhance extraction rate. Our experiments show that the proposed method for extracting perspectively distorted license plates is fairly reliable.

Analysis of 3D Hand Trajectory Gestures Using Stroke-Based Composite Hidden Markov Models

We present a glove-based hand gesture recognition system using hidden Markov models (HMMs) for recognizing the unconstrained 3D trajectory gestures of operators in a remote work environment. A Polhemus sensor attached to a PinchGlove is employed to obtain a sequence of 3D positions of a hand trajectory. The direct use of 3D data provides more naturalness in generating gestures, thereby avoiding some of the constraints usually imposed to prevent performance degradation when trajectory data are projected into a specific 2D plane. We use two kinds of HMMs according to the basic units to be modeled: gesture-based HMM and stroke-based HMM. The decomposition of gestures into more primitive strokes is quite attractive, since reversely concatenating stroke-based HMMs makes it possible to construct a new set of gesture-based HMMs. Any deterioration in performance and reliability arising from decomposition can be remedied by a fine-tuned relearning process for such composite HMMs. We also propose an efficient method of estimating a variable threshold of reliability for an HMM, which is found to be useful in rejecting unreliable patterns. In recognition experiments on 16 types of gestures defined for remote work, the fine-tuned composite HMM achieves the best performance of 96.88% recognition rate and also the highest reliability.

An improved binarization algorithm based on a water flow model for document image with inhomogeneous backgrounds

A segmentation algorithm using a water flow model [Kim et al., Pattern Recognition 35 (2002) 265-277] has already been presented where a document image can be efficiently divided into two regions, characters and background, due to the property of locally adaptive thresholding. However, this method has not decided when to stop the iterative process and required long processing time. Plus, characters on poor contrast backgrounds often fail to be separated successfully. Accordingly, to overcome the above drawbacks to the existing method, the current paper presents an improved approach that includes extraction of regions of interest (ROIs), an automatic stopping criterion, and hierarchical thresholding. Experimental results show that the proposed method can achieve a satisfactory binarization quality, especially for document images with a poor contrast background, and is significantly faster than the existing method.

Preferred skin color reproduction based on adaptive affine transform

This work models a preferred skin color area with an ellipse according to the average luminance of skin and proposes an effective method for skin color reproduction by introducing the affine transform between the skin color ellipse and the preferred skin color ellipse.

Skin Color Detection through Estimation and Conversion of Illuminant Color Under Various Illuminations

Skin color provides a useful cue for detecting faces and reproducing preferred colors. However, skin color detection based on just a static model often decreases the detection rate, as skin color in an image captured by a camera undergoes variations as the illumination changes. Thus, to enhance skin color detection using a static model, the color of an estimated illuminant from images captured under various illumination conditions is converted to that of a canonical illuminant. First, the illuminant color is estimated from the pixels in the sclera region of the eyes, then converted to a canonical color for robust skin color detection. Experimental results show that the proposed skin color detection method increases the detection rate, especially for images taken with a low or high correlated color temperature of illumination.

Driving waveform for reducing temporal dark image sticking in AC plasma display panel based on perceived luminance

Minimizing the reset discharge produced under an MgO-cathode condition and eliminating the wall charges accumulated on the address electrode prior to the reset period are the key factors involved in reducing temporal dark image sticking. Thus, based on the perceived luminance, new driving waveforms that can prohibit an MgO-cathode induced reset discharge or erase the wall charges accumulated on the address electrode prior to the reset period are examined for the complete elimination of temporal dark image sticking without deteriorating the address discharge characteristics. As a result of monitoring the difference in the infrared emission and perceived luminance between the cells with and without image sticking, the proposed driving waveform was shown to effectively remove temporal dark image sticking without deteriorating the address discharge characteristics

Experimental observation of image sticking phenomenon in AC plasma display panel

The itinerant strong sustain discharge that occurs during a sustain period over a few minutes causes image sticking, which means a ghost image remains in the subsequent image when the previous image was continuously displayed over a few minutes. Accordingly, this paper investigates whether the dominant factor in image sticking is the MgO surface or phosphor layer by testing the effects of image sticking in subsequent dark and bright images using a 42-in plasma display panel. When the subsequent image was dark, the image sticking was found to produce a brighter ghost image than the background. Thus, since the luminance of a dark image is produced by the weak discharge that occurs during the reset-period, the higher luminance of the ghost image was mainly due to the activation of the MgO surface. Conversely, when the subsequent image was bright, the image sticking was found to produce a darker ghost image than the background. Thus, since the luminance of a bright image is predominantly produced by the strong discharge that occurs during the sustain period, the lower luminance of the ghost image was mainly due to the deterioration of the phosphor layer.

Experimental observation of temperature- dependent characteristics for temporal dark boundary image sticking in 42-in AC-PDP

The temperature-dependent characteristics of temporal image sticking, especially temporal dark boundary image sticking, are investigated by observing the infrared (IR) emission characteristics relative to the panel-temperature rise during the reset period in a 42-in plasma television. A panel-temperature rise is induced in adjacent cells by the discharge cells (i.e., image-sticking cells) due to the itinerant strong sustain discharge, thereby lowering the firing voltage for both the adjacent and discharge cells during the ramp-up period. However, the phosphor layers in the discharge cells are deteriorated due to the strong discharge, whereas the phosphor layers in the adjacent cells are not degraded due to the absence of the strong sustain discharge. Consequently, when the displayed image is white, the temporal boundary image-sticking cells exhibit a higher luminance with a dark background image. Meanwhile, when a single color image is displayed, the temporal boundary image-sticking cells induce a color difference with a dark background image.

Improvement of color temperature using independent control of red, green, blue luminance in AC plasma display panel

This paper presents a new driving scheme for the improvement and flexibility of a color temperature without sacrificing a peak white luminance using an independent control of the red (R), green (G), and blue (B) luminance in an alternating current plasma display panel (AC-PDP). The independent control for the R, G, and B emissions can be achieved by selective application of the various narrow auxiliary pulses to the R, G, and B address electrodes during a sustain-period. The auxiliary pulses can control the luminance levels independently from the R, G, and B cells by forming the fast and efficient plasma or by slight disturbing of the wall charge accumulation. By the application of various auxiliary pulses leading to the simultaneous control of each colors luminance, it is observed that the new driving scheme can improve the color temperature from 5396 K to 10 980 K in a 4-in test panel with almost the same peak white luminance as that of the conventional driving scheme.

Adaptive window method with sizing vectors for reliable correlation-based target tracking

Abstract We propose an adaptive window method that can provide a tracker with a tight reference window by adaptively adjusting its window size independently into all four side directions for enhancing the reliability of correlation-based image tracking in complex cluttered environments. When the size and shape of a moving object changes in an image, a correlator often accumulates walk-off error. A success of correlation-based tracking depends largely on choosing the suitable window size and position and thus transferring the proper reference image template to the next frame. We generate sizing vectors from the corners and sides, and then decompose the sizing vector from the corner into two corresponding sides. Since our tracker is capable of adjusting a reference image size more properly, stable tracking has been achieved minimizing the influence of complex background and clutters. We tested the performance of our method using 39 artificial image sequences made of 4260 images and 45 real image sequences made of more than 3400 images, and had the satisfactory results for most of them.