Catálogo de publicaciones - libros

In this paper, we propose a method for robustly estimating camera homography using fuzzy RANSAC from the correspondences between consecutive two images. We use a fuzzified version of the original RANSAC algorithm to obtain accurate camera homography in the presence of outliers. The drawback of RANSAC is that its performance depends on a prior knowledge of the outlier scale. To resolve this problem, the proposed method classifies all samples into three classes (good sample set, bad sample set and vague sample set) using fuzzy classification. It then improves classification accuracy omitting outliers by iteratively sampling in only good sample set. Experimental results show the robustness of the proposed approach for computing a homography on real image sequence.

Flashlights in video cause abrupt brightness changes of a scene and will be detected as false scene change if not handled properly. So in this paper propose a robust scene change detection algorithm which can detect the scene change correctly by skipping for the flashing period. At first, the proposed methods make use of histogram comparison which are simple and more robust to object and camera movement while enough spatial information is retained to produce more accurate difference values from consecutive frames. The normalized works of difference values are performed to solve the optimal threshold decision problem. Normalized difference values are dynamically compressed by Log metrics and more efficient to detect scene boundary. Finally, we distinguish flashlights from difference values by applying a ‘flashlights features’ which are defined based on the temporal property of normalized difference values across a frame sequence. The proposed methods are tested on the various video types and experimental results show that the proposed algorithms are effective and reliably detect scene changes.

The Handwrite verification is the technique of distinguishing the same person’s specimen of handwriting from imitations with any two of more handwritten texts using one’s handwritten individuality. The handwrite verification technique is used for distinguishing of ghostwriting and handwritten text or signature verification, criminal identification of handwritten text or signature. This work is subject to a loss of objectivity, may consume too much time to obtain verification, and involves the processing costs of a verification advisor. To solve these problems, we suggest the solution listed above, and automation of the analysis of similarity of texts using a computer pattern analysis. Primal processes of the system which are suggested in this paper are abstraction of letter area, abstraction of feature, study of feature, analysis of studied feature, abstraction of a contrast sample, detection of similarity between two characters, deduction of analyzed results and so forth. It is expected that the system suggested will be widely applicable, and is expected to generate great interest because it will enable both short and long patterns such as signatures and handwriting samples to be processed at the same time.

Internet-based problem solving activities are becoming more prevalent and the resulting learning efficiency gets greater and greater. This paper proposes an easy-to-access active learning system in which information regarding the certificate of qualification, and written and practical evaluation questions are databased. One of the advantages of the proposed system is filtering user-specific questions through the use of information on the user profile to which weight is applied, thereby evaluating learning of individual users, promoting their motivation for learning, and giving them a sense of achievement. Another advantage lies in the fact that the proposed system increases the rate of acquisition of qualification certificates through giving guidance on acquisition of certificates in computer literacy, which will renew learners’ attention and recognition of future career. The proposed system proved to enhance exam performance by up to 10%.

We propose a face and completely facial feature extraction model for facial expression applications. This model applies to both face contour detection and face region detection. First, we introduce skin-color filtering using YCbCr color space to extract the skin-color of face the region. Second, the template ACM is modeled by the active contour model. This model is more active than ASM (Active Shape Model). Our algorithm has been tested in experiments with various subjects, producing a good extraction results.

In this paper, we propose a image retrieval method using skin color feature and shape feature. Section by using a method of snake, consider color information. Section II, image retrieval using by ICSS(Improve Curvature Scale Space). As a result, we show that good results can be obtained by skin color and shape feature.

This paper introduces a method for modification of the formula of the fractal box counting dimension. The method is based on the utilization of theprobability distribution formula in the fractal box count. The purpose of this method is to use it for the discrimination of oil spill areas from the surrounding features e.g., sea surface and look-alikes in RADARSAT-1 SAR data. The result shows that the new formula of the fractal box counting dimension is able to discriminate between oil spills and look-alike areas. The low wind area has the highest fractal dimension peak of 2.9, as compared to the oil slick and the surrounding rough sea. The maximum error standard deviation of low wind area is 0.68 which performs with fractal dimension value of 2.9.

In this paper, we present the development of a simple and low cost data glove system using tilt and flex sensors as a Korean Finger Spelling (KFS) recognition system. This data glove has the capability to measure the palm and finger gesture postures. The process of building a simple KFS recognition system and method for recognizing the KFS letters is also proposed in this paper. The k-means algorithm is used to classify the KFS letter’s based on tilt sensor measurement. The flex sensor measurement on each finger is divided into three main bending positions and quantization index rule-based is used to recognize the KFS letters. For the convenience of using this glove, a simple and efficient calibration process of the finger gesture is provided, so that all the required parameters for recognition can be adapted automatically. The system gives an average of 80% correct recognition for the 24 letters in KFS. The glove-based KFS is possibility to ease and encourage the Korean community to learn KFS by providing hands-on and minds-on learning experiences with an affordable data glove.

In this paper, we present a face tracking and detection algorithm in real time camera input environment. To trace and extract a face image in complicated background and various illuminating conditions, we used pyramidal Lucas-Kanade feature tracker. Also we used KLT algorithm, which has robustness for rotated facial image, to extract the distinguishing feature of face area.

This paper proposes an enhanced snake algorithm using the proximal edge search method. The proposed algorithm adds a new energy term called “proximal edge search” to the existing greedy algorithm without any passive adjustment of weight. The new energy term is represented by the distance between the snake point and the edge when there is a proximal edge. This modified algorithm could improve the accuracy by acquiring the detailed contour of complex objects and actively resolve the passive determination of weight. The validity of the proposed method was proven through experiments.