Joongkyu Kim

Retinex method based on adaptive smoothing for illumination invariant face recognition

In this paper, we propose the Retinex method for illumination invariant face recognition developed on the basis of adaptive smoothing technology. By the well-known Retinex theory, illumination is generally estimated and normalized by smoothing the input image first and then dividing the estimate into the original input image. Therefore, performance mainly depends on how good the estimated illumination is. The proposed method estimates illumination by iteratively convolving the input image with a 3x3 smoothing mask weighted by a coefficient via combining two measures of the illumination discontinuity at each pixel. We address a couple of additional concepts, which are designed to be suitable especially for face images. One is the new conduction function for adaptive smoothing, and the other is the smoothing constraint for more accurate description of real environments. In this way, we can achieve an efficient illumination normalization in which face images with even strong shadows are normalized efficiently. The proposed method is evaluated based on Yale face database B, CMU PIE database and AR face database by applying PCA. The comparative results indicate that the proposed method present consistent and promising results even when images under harsh illumination conditions are used as a training set.

Frizzled 4 Regulates Stemness and Invasiveness of Migrating Glioma Cells Established by Serial Intracranial Transplantation

One of the most detrimental hallmarks of glioblastoma multiforme (GBM) is cellular invasiveness, which is considered a potential cause of tumor recurrence. Infiltrated GBM cells are difficult to completely eradicate surgically and with local therapeutic modalities. Although much effort has focused on understanding the various mechanisms controlling GBM invasiveness, its nature remains poorly understood. In this study, we established highly serial intracranial transplantation. U87R4 cells were highly invasive and displayed stem cell-like properties, as compared to noninvasive but proliferative U87L4 cells. Microarray analysis during serial transplantation revealed that apoptosis-inducing genes (caspase3 and PDCD4) were downregulated whereas several cancer stem cell-relevant genes [Frizzled 4 (FZD4) and CD44] were upregulated in more invasive cells. U87R4 cells were resistant to anticancer drug-induced cell death, partly due to downregulation of caspase3 and PDCD4, and they retained activated Wnt/β-catenin signaling due to upregulation of Frizzled 4, which was sufficient to control neurosphere formation. We also found that FZD4 promoted expression of the epithelial to mesenchymal transition regulator SNAI1, along with acquisition of a mesenchymal phenotype. Taken together, our results argue that Frizzled 4 is a member of the Wnt signaling family that governs both stemness and invasiveness of glioma stem cells, and that it may be a major cause of GBM recurrence and poor prognosis.

A stroke filter and its application to text localization

Most researchers have used edge, intensity, corner, and texture features for text localization in video images. However, these features do not fully coincide with the features of the text, and can not fulfill all the necessary conditions of the text. Therefore, it is very difficult to localize text robustly in video images which have complex backgrounds with strong edge or texture clutter using these features. In this paper, we propose a stroke filter which can detect strokes of texts for robust text localization. By using this stroke filter, we can remove text candidates which have strong edges but are not text. Furthermore, we apply the stroke filter to our text localization system and localize text more robustly in the video images. The effectiveness and efficiency of the proposed method is verified by extensive experiments on a challenging database containing 480 video images.

Depth-image-based rendering for 3DTV service over T-DMB

In this paper, we present a depth-image-based rendering (DIBR) technique for three-dimensional television (3DTV) service over terrestrial-digital multimedia broadcasting (T-DMB). 3DTV service over T-DMB is very attractive because the single user environment of T-DMB is suitable to glassless 3D viewing. However, the bit budget for the transmission of additional data in 3DTV service is very limited in T-DMB. DIBR can resolve this problem because the depth information as the additional data can be compressed effectively. However, DIBR has also some problems such as the backward compatibility with the conventional T-DMB, a large computational cost for rendering virtual images and generation of holes by disocclusion. To solve these problems, we first present the service architecture for DIBR-based 3DTV service over T-DMB, which can maintain the backward compatibility with the conventional T-DMB. Second, we propose a new depth preprocessing method based on adaptive smoothing. In the proposed depth preprocessing method, two adaptive smoothing filters of the discontinuity-preserving smoothing and the gradient direction-based smoothing are sequentially conducted. As a result, both the reduction of bitrates required for the depth transmission and the reduction of the holes can be achieved. Finally, we present look-up table (LUT)-based simultaneous method for generating an auto-stereoscopic image so that problems of the limited memory and the large computational cost can be resolved in T-DMB player. Various experiments show that the proposed DIBR technique can be efficiently employed for 3DTV service over T-DMB.

A Structure for 2D/3D Mixed Service Based on Terrestrial DMB System

This paper introduces a structure of delivering 2D/3D mixed service including additional 3D images as well as 2D images through terrestrial DMB system. T-DMB can provide video service and data service within the conventional T-DMB standards for video and data stream transmission. To provide the 2D/3D mixed service, it is necessary to define the methodology for handling 3D additional data stream with keeping backward compatibility with 2D service. Therefore we suggest the structure to adapt 3D additional data to conventional T-DMB structure and show its prototype implementation that provides 2D/3D mixed service.

Accurate text localization in images based on SVM output scores

In this paper, we propose a new approach for accurate text localization in images based on SVM (support vector machine) output scores. In general, SVM output scores for the verification of text candidates provide a measure of the closeness to the text. Up to the present, most researchers used the score for verifying the text candidate region whether it is text or not. However, we use the output score for refining the initial localized text lines and selecting the best localization result from the different pyramid levels. By means of the proposed approach, we can obtain more accurate text localization results. Our method has three modules: (1) text candidate detection based on edge-CCA (connected component analysis), (2) text candidate verification based on the classifier fusion of N-gray (normalized gray intensity) and CGV (constant gradient variance), and (3) text line refinement based on the SVM output score, color distribution and prior geometric knowledge. By means of experiments on a large news database, we demonstrate that our method achieves impressive performance with respect to the accuracy, robustness and efficiency.

Fast communication: A new approach for text segmentation using a stroke filter

We propose a new method for achieving robust text segmentation in images by using a stroke filter. It is known that to segment text accurately and robustly from a complex background is a very difficult task. Most of the existing methods are sensitive to text color, size, font, and background clutter, because they use simple segmentation methods or require prior knowledge about text shape. In this paper, we attempt to consider the intrinsic characteristics of the text by using the stroke filter and design a new and robust algorithm for text segmentation. First, we describe the stroke filter briefly based on local region analysis. Second, the determination of text color polarity and local region growing procedures are performed successively based on the response of the stroke filter. Finally, the feedback procedure by the recognition score from an optical character recognition (OCR) module is used to improve the performance of text segmentation. By means of experiments on a large database, we demonstrate that the performance of our method is quite impressive from the viewpoints of the accuracy and robustness.

Depth Image Based Rendering for 3D Data Service Over T-DMB

In this paper, we present a depth image based rendering technique for 3D data service over T-DMB. 3D service over T-DMB is very attractive because the single user environment of T-DMB is suitable to glassless 3D viewing. However, the bit budget for depth transmission over T-DMB is very limited with 32 Kbps for data service. Depth image based rendering can resolve this problem because the corresponding depth sequence can be compressed effectively. However, depth image based rendering has also some problems such as a large computational cost for generating virtual images and generation of holes by disocclusion. Therefore, we propose the new depth preprocessing method based on adaptive smoothing and the simultaneous method for generating an auto-stereoscopic image. In the proposed depth preprocessing method, the discontinuity preserving smoothing, followed by the adaptive smoothing based on gradient direction, is conducted by generalizing the conduction function of the anisotropic smoothing. As a result, the reduction of both bit rate required for transmission and holes is achieved. Additionally, the simultaneous method for generating an auto-stereoscopic image can resolve problems of limited memory and large computational complexity. Some experiments show that the proposed scheme can be efficiently employed for 3D data service over T-DMB.

Player Information Extraction for Semantic Annotation in Golf Videos

In sports videos, text provides semantic information about the game such as scores and players. This paper provides an accurate extraction method of the player information in golf. First, a new detection method of the key captions containing the player information is presented. Since the location of the key captions containing the player information is not fixed during a game in golf, we use a color pattern of captions and its temporal repetition property instead of the location property to decide the key captions. Second, a dual binarization method is presented to segment texts with different color polarities (i.e. dark and bright texts) easily from the background in the key captions. Finally, the binarization results are recognized by OCR and converted to plain texts. The player is recognized by comparing the plain texts with the pre-reserved player name database. Experiments on a large database show that our method can extract the player information efficiently in golf videos.

2D/3D Mixed Service in T-DMB System Using Depth Image Based Rendering

In this paper, we introduce a 2D/3D mixed service in terrestrial digital multimedia broadcasting (T-DMB) system using depth-image-based rendering (DIBR). The 2D/3D mixed service is the 3D service type that 3D contents are shown partially while a 2D video sequence is displayed in the entire screen, or vice versa. This service is very attractive because partial display of 3D contents can maximize the effect of 3D services and can reduce the viewers fatigue while watching a 3D display. Especially, we can provide more flexible 3D services with a low amount of data by using DIBR. Moreover, 3D contents can be manipulated efficiently by XML-based metadata. The proposed 2D/3D mixed service is tested using various contents in the 3D display with parallax barrier. The proposed 2D/3D mixed service can be employed efficiently as the additional data services in conventional T-DMB structure.