Younglae Bae

An effective method for combining multiple features of image retrieval

We propose an image retrieval system using the Borda count method in combining multiple experts (CME). It combines color-, shape- and texture-based retrieval systems. CME method can complementarily combine results of each retrieval system, which uses different features. There are some problems when the Borda count method in pattern recognition is applied to image retrieval. Thus, we propose a modified Borda count method to solve these problems. In the experiment our method reduces false positive errors and produces better results than that of each retrieval module that uses only one feature.

Automatic text extraction in digital videos using FFT and neural network

Popular methods for extracting a text region in video images are in general based on analysis of a whole image such as merge and split method, and comparison of two frames. Thus, they take long computing time due to the use of a whole image. Therefore, this paper suggests the faster method of extracting a text region without processing a whole image. The proposed method uses line sampling methods, FFT and neural networks in order to extract texts in real time. In general, text areas are found in the higher frequency domain, thus, can be characterized using FFT. The candidate text areas can be thus found by applying the higher frequency characteristics to neural network. Therefore, the final text area is extracted by verifying the candidate areas. Experimental results show a perfect candidate extraction rate and about 92% text extraction rate. The strength of the proposed algorithm is its simplicity, real-time processing by not processing the entire image, and fast skipping of the images that do not contain a text.

A method for original image recovery for caption areas in video

It is sometimes necessary to remove the captions from video images already broadcast and recover the original images. Traditional research on image restoration has focused on restoring blurred images to sharp images using frequency filtering. This paper proposes a method for extracting caption regions and a method for automatically recovering original. The caption region detection method extracts caption candidate regions using TF (topographic feature) of characters, and verifies candidate regions to determine the final caption regions. From the result of caption region detection, we know the start and end frames of captions in video and the character areas in the caption regions. For original image recovery, we use information such as the start and end frames of the detected caption, camera operation, scene change and motion vector of extracted characters in caption regions. The direction for the recovery is decided using information on the scene change, the start and end frame for captions. According to the direction, we recover the original image using information of camera operation and Block Matching Algorithm (BMA) for character components in extracted caption region. Experimental results show that the case of stationary images having little motions is well recovered. We see that the case of images having motion in complex background is also recovered.

Efficient video images retrieval by using local co-occurrence matrix texture features and normalised correlation

Since multimedia data is widely used in many fields, research is needed on how to search that data. As a method of retrieving multimedia data, especially for image data, content based image retrieval (CBIR) systems have obtained lots of interest. In order to develop CBIR systems, many different properties of an image have been considered for methods of retrieving image data such as colour, texture, shape, etc. The paper proposes a simple and efficient image retrieval algorithm, which retrieves images on the basis of subregional texture characteristics. In order to retrieve images in terms of contents, it is in general required to obtain a precise segmentation. However, it is very difficult and takes a lot of computing time for precise segmentation. Therefore, the proposed algorithm subdivides an image in terms of a certain sized windows, and finds out whether an image contains a query pattern on the basis texture features. Also, these texture features are described by only six different texture features produced from a co-occurrence matrix. For an image retrieval system, a normalised correlation is adopted as a similarity function, which is not dependent on the range of texture feature values. Finally, the proposed algorithm is applied to various images and produces competitive results.

Text extraction in videos using topographical features of characters

Traditional character extraction methods use split-and-merge or color segmentation methods from the whole image. These methods do not reflect characteristics of characters such that character regions can be extracted incorrectly. In this paper, we use topographical features of characters to extract the character points and use the density of those points to extract the candidate regions for captions. Character regions are determined by testing several conditions and verifying those candidate regions. Experimental results show that the candidate region extraction rate is 100%, and the character region extraction rate is 99%. Non-character regions can exist within the extracted character regions which are removed using the Isodata color clustering method.

Efficiency Analysis of Schisandra Tea Using Image & Acoustic Signal Processing

Abstract We propose an efficiency analysis method to test whether Schisandra tea is beneficial to the health of internal organs, and, if so, to find which organ is mostly benefited. Firstly, for this, the color change in the right cheek region of face images before and after drinking Schisandra tea is analyzed using image processing techniques. Also, acoustic analysis experiments on both of the vocal cords vibration and the intensity value of acoustic signal are performed to investigate further Schisandra teas effectiveness. The results of lung function smoothly by eating omijacha become. Vocal cords vibration and voice energy intensity of the right cheek region has changed a steady increase of the b values were observed. Key Words : Image Analysis, Acoustic Analysis, Schisandra Tea, Right Cheek, Vocal Cords, Intensity, * 교신저자 : 조동욱(ducho@cpu.ac.kr)접수일 11년 05월 06일 수정일 (1차 11년 06월 08일, 2차 11년 06월 13일) 게재확정일 11년 07월 07일 1. 서론 현재 우리나라는 OECD 국가 중 평균 수명은 높은 편이지만 건강 수명은 66세로 가장 낮은 편이다. 이를 해결하기 위해서는 질병의 치료보다는 예방과 보건, 약품보다는 식품을 통해 우리의 건강 수명을 높이는 것에 대한 방안이 적극 강구되어야 하는 실정이다. 통상 우리의 일상생활에서 건강 수명을 높이기 위한 방법으로는 음식 섭취와 규칙적인 운동의 방법이 사용되어진다. 또한 이를 위한 다양한 운동법이나 건강 보조식품 등이 나오고 있는 것도 정확한 현 실정이다. 그런데 현재 중요한 것 중 하나가 건강 중진에 도움이 된다고 하여 즐겨 먹는 건강 보조 식품 등이 실제 건강 증진에 도움이 되느냐 하는 것이다. 그러나 현재까지 이를 위한 과학적, 객관적, 정략적, 시각화된 결과가 전무한 실정이라는 사실이다. 즉, 몸에 좋다고 하여 마시기는 하지만 이를 규명할 과학적 자료가 현재까지 존재하지 않았다. 이를 위해 본 논문에서는 IT 기술을 적용하여 우리들이 건강 증진을 위해 가장 많이 섭취하는 차 종류의 음식물이 실제 효과가 있는 것인지에 대한 분석을 행하고자 한다. 그 중 본 논문에서는 우선적으로 오미자차를 연구 대상으로 잡았다. 오미자라는 이름이 단맛, 신맛, 매운맛, 짠맛, 쓴맛을 갖고 있다고

A Proposal of Sasang Constitution Classification in Middle-aged Women Using Image and Voice Signals Process

Sasang medicine is our country`s unique traditional medicine based on the classification of individual physical constitution. In the Sasang medicine, what is considered the most important task is to categorize Sasang constitution exactly. Therefore, security of objective elements and diagnosis index is a problem awaiting solution in Sasang constitution classification. To this the paper abstracted result value from objectification, visualization, a fixed quantity of Sasang constitution to analyze face image signals and voice signals. And, comparing the differences constitution would like to develop system classification of Sasang constitution. Specially, image and voice signals are different because of gender, age, region so it composed Sasang constitution group to 40-50 years women in Seoul. To extract of these image and voice signals wanted to perform comparison, analysis in constitution. Finally, it would like to prove a significance of research result through experiment.

A fast retrieval method for image features

With a very large size image database, there should be a way to search for the similar images efficiently. However, the traditional approaches that have been adopted to retrieve text or scalar values cannot be used because of the multi-dimensionality of the image feature vectors. A new retrieval method that uses one or more scalar values induced from the image vector is proposed. In the proposed method, the scalar values are independently stored as indices to the target feature vectors. With the proposed method, the searching speed is improved approximately 25 times compared to the sequential searching.

Reconstruction in an Original Video Sequence of the Lost Region Occupied by Caption Areas

Conventional researches on image restoration have focused on restoring blurred images to sharp images using frequency filtering or video coding for transferring images. In this paper, we proposes a method for recovering original images using camera motion and video information such as caption regions and scene changes. The method decides the direction of recovery using the caption information and scene change information. According to direction of recovery, a rough estimate of the direction and position of the original image is obtained using calculated motion vector from camera motion. Because the camera motion dose not reflects local object motion, some distortion can happen in the recovered image. To solve this problem, block matching algorithm that is applied in units of caption character components on the obtained recovery positions. Experimental results show that the case of images having little motions is well recovered. We see that the case of images having motion in complex background is also recovered.

Entropy and color correlation for image indexing

This paper presents a new indexing system, which divides an image into higher and lower entropy regions, and calculates the color correlation features between each type of region. Thus, we can improve the discrimination power of color indexing techniques. The indexing system proposed in this paper has three important properties. The first is an entropy feature. An image is divided by an entropy feature into two regions, lower and higher entropy regions in order to avoid the problems caused by general global features. The second is a color correlation feature. This paper uses 2-dimensional probability distribution functions of an image to obtain color moment features, and thus, obtain more information than using 1-dimensional probability distribution functions (i.e. histogram). The last is the retrieval procedure consisting of two steps: firstly, a simple retrieval algorithm is applied to all the images in a database, and secondly, only the results of the previous retrieval are searched. Thus, it can help in reducing the total retrieval time and improving the retrieval accuracy.