Damon Shing-Min Liu

An intelligent clothes search system based on fashion styles

This work presents an intelligent clothes search system based on domain knowledge, targeted at creating a virtual assistant to search clothes matched to fashion and userpsila expectation using all what have already been in real closet. All what garment essentials and fashion knowledge are from visual images. Users can simply submit the desired image keywords, such as elegant, sporty, casual, and so on, and occasion type, such as formal meeting, outdoor dating, and so on, to the system. And then the fashion style recognition module is activated to search the desired clothes within the personal garment database. Category learning with supervised neural networking is applied to cluster garments into different impression groups. The input stimuli of the neural network are three sensations, warmness, loudness, and softness, which are transformed from the physical garment essentials like major color tone, print type, and fabric material. The system aims to provide such an intelligent user-centric services system functions as a personal fashion advisor.

A user-oriented ontology-based approach for network intrusion detection

A new approach is suggested for designing and developing an intrusion detection application where the domain expertise is used for generating it more easily. This approach uses ontologies as a way of grasping the knowledge of a domain, expressing the intrusion detection system much more in terms of the end users domain, generating the intrusion detection more easily and performing intelligent reasoning. Furthermore, ontologies are also used as a conceptual modeling tool allowing a non-expert person to model his/her intrusion detection application only using the concepts of intrusion detection more intuitively and more user-oriented towards his/her domain expertise.

Use of hardware Z-buffered rasterization to accelerate ray tracing

Ray tracing is a rendering technique for producing realistic 3D computer graphics. Compared to traditional scan-line rendering which is generally adopted by graphics pipeline, ray tracing can simulate more realistic global illumination, however, with the cost of expensive computation. In this paper, we implement a ray tracer that combines advantages of both rendering schemes: efficiency of scan-line rendering and reality of ray tracing. We first use hardware-accelerated rasterization with Z-buffer to quickly determine the first ray-triangle hit of eye rays on the GPU. Secondary rays such as reflective and shadow rays are then traced to generate global illumination on the CPU with a bounding volume hierarchy (BVH) which plays the role of our acceleration structure. The experiments show that rasterization is much more efficient in finding the first hit and can completely replace the traditional ray casting procedure.

A Decision Making Framework for Dressing Consultant

The project, Dressing Consultant, aims to provide a system which functions as a personal wearing advisor to help general users choose a correct clothing for occasions. ALCOVE (attention learning covering network) neural network model is used to train the matchmaker as a fashion editor. In addition, image processing techniques are employed at pre-processing stage to obtain the essential data of garments and to build a digital wardrobe for individuals. On the occasions when user has trouble finding an outfit for a special event, what user could do is to make a decision of the style of apparel to the system and let the system go through piece of garments in the digital wardrobe, and the matchmaker will then find several matched pairs. Eventually, the most similarly suitable and matched garments pair is shown in 3D show room. This paper focuses on making decision of correct clothing according to those classifying and matching rules extracted from fashion industry

The Problem of Time Scales in Computer Visualization

A new approach to temporal aspects of visualization is proposed. Empirical and theoretical studies of visualization processes are presented in a more general context of human-computer interaction (HCI). It makes possible to develop a new model of HCI as a base for visualization. This model implements a methodology of autonomous intelligent agents. Temporal aspects of the model are rooted in the system theory of time. This theoretical background provides for obtaining properties of time coordination. In particular, it is proved that some theoretically advanced models for computation and visualization cannot be realized by physical devices. Applications of the model, its implications for visualization and problems of realization are considered. The new approach does not refute or eliminate previous research in this area, but puts it in a more general and adequate context, systematizing it and providing efficient facilities for the development of computer visualization.

A User-centric Intrusion Detection System by Using Ontology Approach

In the security infrastructure, intrusion detection has become an indispensable defense line in face of increasing vulnerabilities exposed in today’s computing systems and Internet. In this paper, our approach uses ontologies as a way of grasping the knowledge of a domain, expressing the intrusion detection system much more in terms of the end users domain, generating the intrusion detection more easily and performing intelligent reasoning. Experimental results show that our anomaly detection techniques are very promising and are successful in automatically detecting intrusions at very low false alarm rate compared with several important traditional classification techniques.

Using prefetching to improve walkthrough latency

Object correlations are common semantic patterns in walkthrough (WT) systems. They can be exploited for improving the effectiveness of storage caching, prefecthing, data layout, and minimization of query‐response times. Previous approaches for reducing I/O access time are seldom investigated. On the other side, data mining techniques extract implicit, previously unknown and potentially useful information from the databases. However, those methods are presented for typical data mining datasets and not suitable for our WT system datasets. This paper proposes a class of novel and efficient pattern‐growth method for mining various frequent sequential traversal patterns in the WT. Our pattern‐growth method adopts a divide‐and‐conquer approach to decompose both the mining tasks and the databases. The frequent sequential traversal patterns are used to predict the user navigation behavior and help to reduce disk access time with proper placement patterns into disk blocks. We also define the terminologies such as paths, views, and objects used in the model. We have done extensive experiments to demonstrate how these proposed techniques not only significantly cut down disk access time, but also enhance the accuracy of data prefetching. Copyright

A Physics-Based Augmented Reality Jenga Stacking Game

In this paper, we present a cross-platform environment in which handheld and desktop-computer users can collaborate with each other in a shared scene to accomplish physically realistic experiences during the course of interaction. In realizing the system, we: 1) exploit client/server architecture to connect user devices, where the server is responsible for maintaining and managing the virtual objects shared by users, 2) build the distributed scene graph structure according to X3D specification for presenting 3D computer graphics, 3) use mobile phones as tangible input object to provide an intuitive interaction interface, 4) design a natural hand interface which enables users to directly manipulate virtual object using their bare hands. To demonstrate the effectiveness of the system, we developed Jeng AR, a simulation of the well-known tower building game.

Discovering Dressing Knowledge for an Intelligent Dressing Advising System

Our research aims to develop a system to help women choose correct attire for attending a specific occasion using all of what have already been in their closets. Many different computer theories and techniques are gathered in the project. Category learning with supervised neural networking is applied to cluster garments into different impression groups. Fuzzy theories are applied for gathering fashion match rules. In addition, modeling and virtual dressing techniques are used for representing matched garments pair in digital show room. User can simply submit her queries to the system on the occasions when the user has trouble finding an outfit for a special event. After enquiries are received, the core is following fuzzy logic rules to search good matches in the garment database and showing the matched results in the show room. This paper focuses on how garment classification and matching rules retrieved from fashion stylists.

PrefixUnion: Mining Traversal Patterns Efficiently in Virtual Environments

Sequential pattern mining is an important data mining problem with broad applications. Especially, it is also an interesting problem in virtual environments. In this paper, we propose a projection-based, sequential patterngrowth approach, called PrefixUnion. Meanwhile, we also introduce the relationships among transactions, views and objects. According to these relationships, we suggest two mining criteria — inter-pattern growth and intra-pattern growth, which utilize these characteristics to offer ordered growth and reduced projected database. As a result, the large-scale VRML models could be accessed more efficiently, allowing for a real-time walk-through in the scene.