Phill Kyu Rhee

Web personalization expert with combining collaborative filtering and association rule mining technique

Abstract Web personalization has been providing electronic businesses with ways to keep existing customers and to obtain new ones. There are two approaches for providing personalized service: a content-based approach and a collaborative filtering approach. In the content-based approach, it is not easily applied to web objects (pages, images, sounds, etc) which are represented by multimedia data type information. Collaborative filtering approaches have cold-start problem. More serious weakness of collaborative filtering is that rating schemes can only be applied to homogenous domain information. In this paper, we present a framework of personalization expert by combining collaborative filtering method and association rule mining technique to overcome problems that traditional personalized systems have. Since multimedia data type web object cannot be easily analyzed, we adopted a collaborative filtering method that considers each object as an item, and attempts a personalized service. Similar users of each domain object are found as the result of the collaborative filtering method. These similar users’ web object access data is used by apriori algorithm to discover object association rules.

Color Based Hand and Finger Detection Technology for User Interaction

The aim of this paper is to present the methodology for hand detection and propose the finger detection method. The detected hand and finger can be used to implement the non-contact mouse. This technology can be used to control the home devices such as curtain and television. Skin color is used to segment the hand region from background and counter is extracted from the segmented hand. Analysis of counter gives us the location of finger tip in the hand. We have performed extensive experiment and achieve very encouraging result.

Embedded face recognition based on fast genetic algorithm for intelligent digital photography

In this paper, we propose embedded face recognition (FR) to use in intelligent image system. For efficient FR VLSI design, we use a feature selection and feature extraction method based on Gabor wavelets using a fast genetic algorithm (FGA). Many FR systems are based on Gabor wavelet due to its desirable characteristics of spatial locality and orientation selectivity. However, the process of searching for features with Gabor wavelet is computationally expensive and has an unusual sensibility for variations such as illumination. To overcome these problems and use in real-time applications, we optimize Gabor wavelets parameters of translation, orientations and scales, which make it approximates a local image contour region by the use of hardware oriented FGA. From experimental results, we certify that our method shows recognition rate of over 97.27 % for FERET dataset, which exceeds the performance of the other popular methods

Multi-class Multi-object Tracking Using Changing Point Detection

This paper presents a robust multi-class multi-object tracking (MCMOT) formulated by a Bayesian filtering framework. Multi-object tracking for unlimited object classes is conducted by combining detection responses and changing point detection (CPD) algorithm. The CPD model is used to observe abrupt or abnormal changes due to a drift and an occlusion based spatiotemporal characteristics of track states. The ensemble of convolutional neural network (CNN) based object detector and Lucas-Kanede Tracker (KLT) based motion detector is employed to compute the likelihoods of foreground regions as the detection responses of different object classes. Extensive experiments are performed using lately introduced challenging benchmark videos; ImageNet VID and MOT benchmark dataset. The comparison to state-of-the-art video tracking techniques shows very encouraging results.

Real time head nod and shake detection using HMMs

This paper discusses a technique of detecting a head nod and shake. The proposed system is composed of face detection, eye detection and head nod and head shake detection. We use motion segmentation algorithm that makes use of differencing to detect moving peoples faces. The novelty of this paper comes from the differencing in real time input images, preprocessing to remove noises (morphological operator and so on), detecting edge lines and restoration, finding the face area and cutting the head candidate. Moreover, we adopt K-means algorithm for finding head. Eye detection extracts the location of eyes from the detected face region. It is performed at the region close to a pair of eyes for real-time eye detecting. Head nod and shake can be detected by HMMs those are adapted by a directional vector. The HMMs vector can also be used to determine neutral as well as head nod and head shake. These techniques are implemented on a lot of images and a notable success is notified.

Optimal Gabor Encoding Scheme for Face Recognition Using Genetic Algorithm

This paper describes methods that optimize Gabor wavelet encoding scheme using Genetic algorism. Gabor wavelet is known very effective that extract important characteristic in object recognition. This paper presents, using the Genetic algorithm, an optimization methodology of the Gabor encoding scheme so that it produces characteristic vectors effective for the object recognition task. Most previous object recognition approaches using Gabor wavelet do not include careful and systematic optimization of the design parameters for the Gabor kernel, even though the system might be much sensitive to the characteristics of the Gabor encoding scheme. Purpose of this paper investigates geometrical position of Gabor Encode schema and fiducial points for efficient object recognition. Face images in the class of well-defined image objects are used. The superiority of the proposed system is shown using IT-Lab and FERET. The experiment performed with the proposed system exceeds those of most popular methods.

A Novel Image Preprocessing by Evolvable Neural Network

This paper presents a novel and efficient preprocessing method to relieve the effect of changing illumination by restructuring itself under dynamically changing environments. It monitors situations and evolves its structure accordingly, stores its experiences in the form of artificial chromosomes. It performs adaptive preprocessing by reorganizing its structure using the knowledge in the chromosomes matched to an operation environment. Introducing the concept of combining situation-awareness using the evolvable neural network and the evolutionary computing using the Genetic algorithm, the proposed method not only achieves highly efficient preprocessing for object recognition in varying illumination environments, but also solves the time-consuming problem of the evolutionary computing method. The proposed method has been tested and applied successfully to the preprocessing of face images. Face images are in spacially well-defined object class, and the features of face images are represented by multiple fiducial points, each of which is described by the Gabor wavelet transform. The superiority of the proposed preprocessing method is proven by showing the improvements of object recognition accuracy of face dataset: our lab, the AR, and the Yale.

Boundary extraction of moving objects from image sequence

In this paper we address a new approach of contour extraction of moving objects using active contours in an image sequence. Both the spatial convolution process and the external boundary tracing algorithm (EBTA) are used to resolve the problem of automatically locating the initial contour for the external boundaries of an object in the image. Ongoing directions where the contour moves are decided by one of three directions instead of eight directions. We could reduce the time complexity of the energy minimizing operation of the contour by considering only three ongoing directions. In order to apply the proposed algorithm to consecutive imagery, a Kalman filter, an optimal parameter estimator, is adopted. The algorithm extracts the contour of an object from only two consecutive images and takes 0.36 seconds of CPU time.

Adaptive feature representation for robust face recognition using context-aware approach

In this paper, we investigate how to preprocess and accurately identify features from both normal and bad input images, for robust face recognition under uneven illumination environments. Bad illumination is the most challenging problem when implementing robust face recognition systems. From extensive experiments, we found that the performance of individual filtering methods for image enhancement is highly dependent upon application environments. For example, retinex provides good performance under bad illumination. However, it provides very poor performance under normal illumination. On the other hand, histogram equalization provides sufficiently good performance for normal images, however, performance falls dramatically under bad illumination. Since no prior knowledge of the system-working environment can be assumed, the proposed method tries to provide adaptive preprocessing as well as feature representation configuration by exploring the filter combination based on illumination context-awareness. The proposed method has been tested on Inha and FERET image datasets at the preprocessing and feature representation stages for robust face recognition under uneven illumination. Extensive experiments show the proposed system can achieve exceptional performance in varying illumination environments.

Document ontology based personalized filtering system (poster session)

We propose the use of the personalized ontology model to improve the effectiveness of web documents filtering process. One important feature of this model is that by constructing the user specific ontology, web documents can be classified by using the user oriented meta data that reflects the users view about the documents concept. Another is that by applying the user model to searching the classified documents, we achieved the effective document search performance. To find the users preference, Bayesian Learner accepts users interests flow as an input and writes output to users profile. Based on those user profiles, user specific ontologies are constructed to provide efficient search environment.